US RE42728 E1
Interactive interfaces to video information provide a displayed view of a quasi-object called a root image. The root image consists of a plurality of basic frames selected from the video information, arranged such that their respective x and y directions are aligned with the x and y directions in the root image and the z direction in the root image corresponds to time, such that base frames are spaced apart in the z direction of the root image in accordance with their time separation. The displayed view of the root image changes in accordance with a designated viewing position, as if the root image were a three-dimensional object. The user can manipulate the displayed image by designating different viewing positions, selecting portions of the video information for playback and by special effects, such as cutting open the quasi-object for a better view. A toolkit permits interface designers to design such interfaces, notably so as to control the types of interaction which will be possible between the interface and an end user. Implementations of the interfaces including editors and viewers are also disclosed.
1. A method of delivering video over a network, comprising:
receiving video data representing a video sequence at a first device;
generating a hyper-media container containingincluding data associated with the video data;
storing the video data in a memory at the first device;
storing the hyper-media container in the memory in a primary storage format; and
providing the video data and the hyper-media container available over the network to a remote usersecond device, the hyper-media container being provided in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the second device,
wherein the hyper-media container in the secondary storage format includes address information of annotation data.
2. The method as set forth in
3. The method as set forth in
4. The method as set forth in
5. The method as set forth in claim 4 55, wherein controlling access comprisesincludes controlling access to annotations.
6. The method as set forth in claim 4 55, wherein controlling access comprisesincludes controlling access to annotation packs.
7. The method as set forth in claim 4 55, wherein controlling access comprisesincludes controlling access to versions of annotations.
8. The method as set forth in
9. The method as set forth in
10. The method as set forth in
11. The method as set forth in
12. The method as set forth in
13. The method as set forth in
14. The method as set forth in
15. The method as set forth in
16. The method as set forth in
17. The method as set forth in
18. The method as set forth in
19. The method as set forth in
20. The method as set forth in
21. The method as set forth in
22. The method as set forth in
23. The method as set forth in
24. A method of delivering video over a network, comprising:
receiving video data representing a video sequence at a first device;
generating a hyper-media container containingincluding data associated with the video data;
storing the video data in a memory at the first device;
storing the hyper-media container in the memory in a primary storage format; and
providing the video data and the hyper-media container available over the network to a remote usersecond device, the hyper-media container being provided in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the second device,
wherein generating the hyper-media container includes analyzing the video data and associating results of the analyzing with the hyper-media container, and the hyper-media container in the secondary storage format includes address information of annotation data.
25. The method as set forth in
26. The method as set forth in
27. The method as set forth in
28. The methodsmethod as set forth in
29. The method as set forth in
30. The method as set forth in
31. The method as set forth in
32. A method of delivering video over a network, comprising:
receiving video data representing a video sequence at a first device;
generating a hyper-media container containingincluding data associated with the video data;
storing the video data in a memory at the first device;
storing the hyper-media container in the memory in a primary storage format;
providing the video data and the hyper-media container available over the network to a remote usersecond device, the hyper-media container being provided in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the second device;
the method further comprising:
receiving modifications to one of the hyper-media container and the video data from the remote usersecond device and modifying the corresponding one of the hyper-media container and video data; and
publishing versions of the modifications from the remote usersecond device to other remote usersdevices,
wherein the hyper-media container in the secondary storage format includes address information of annotation data.
33. A method of providing video data over a network comprising:
receiving video data representing a video sequence at a first device;
generating a hyper-media container including data associated with the video data;
storing the video data in a memory at the first device;
storing the hyper-media container in the memory in a primary storage format; and
providing the video data and the hyper-media container available over the network to a second device, the hyper-media container being provided in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the second device,
wherein the hyper-media container is separated from the video data and the hyper-media container in the secondary storage format includes address information of annotation data.
34. The method as set forth in
35. The method as set forth in
36. The method as set forth in
37. The method as set forth in
38. The method as set forth in
39. The method as set forth in
40. The method as set forth in
41. The method as set forth in
42. The method as set forth in
43. The method as set forth in
44. A method of providing annotation data over a network comprising:
receiving video data representing a video sequence at a first device;
generating annotation data including data associated with the video data;
storing the annotation data in a memory at the first device; and
providing the digital data and annotation data available over the network to a second device, wherein the annotation data is separated from the video data and is associated with the video data through an identifier,
wherein the annotation data is associated with a hyper-media container, the hyper-media container is stored at the first device in a primary storage format and provided to the second device in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the second device, and the hyper-media container in the secondary storage format includes address information of the annotation data.
45. The method as set forth in
46. The method as set forth in
47. The method as set forth in
48. The method as set forth in
49. The method as set forth in
50. The method as set forth in
51. The method as set forth in
52. The method as set forth in
53. The method as set forth in
54. The method as set forth in
55. The method as set forth in
56. The method as set forth in
57. The method as set forth in
58. An apparatus for delivering video over a network, comprising:
a receiving unit configured to receive video data representing a video sequence;
a generating unit configured to generate a hyper-media container including data associated with the video data;
a memory unit which stores the video data and the hyper-media container, the hyper-media container being stored in a primary storage format;
a providing unit configured to provide the video data and the hyper-media container available over the network to a first device, the hyper-media container being provided in a secondary storage format, the secondary storage format being a format different than the primary storage format and being a format that is readable at the first device,
wherein the hyper-media container in the secondary storage format includes address information of annotation data.
This application is CIP of application Ser. No. 08/887,992; filed on Jul. 3, 1997, now U.S. Pat. No. 5,963,203.
The present invention relates to network distribution and management of video and, more particularly, to distribution and management of interactive video and multi-media containers. The network distribution and management includes, but is not limited to, managing media files, creating and authoring media containers, publishing and indexing media containers, searching and browsing media containers, and distributing media containers.
Video information is being produced at an ever-increasing rate and video sequences, especially short sequences, are increasingly being used, for example, in websites and on CD-ROM, and being created, for example, by domestic use of camcorders. There is a growing need for tools enabling the indexing, handling and interaction with video data. It is particularly necessary for interfaces to be provided which enable a user to access video information selectively and to interact with that information, especially in a non-sequential way.
Conventionally, video information consists of a sequence of frames recorded at a fixed time interval. In the case of classic television signals, for example, the video information consists of 25 or 30 frames per second. Each frame is meaningful since it corresponds to an image which can be viewed. A frame may be made up of a number of interlaced fields, but this is not obligatory as is seen from more recently proposed video formats, such as those intended for high definition television. Frames describe the temporal decomposition of the video image information. Each frame contains image information structured in terms of lines and pixels, which represent the spatial decomposition of the video.
In the present document, the terms “video information” or “video sequences” refer to data representing a visual image recorded over a given time period, without reference to the length of that time period or the structure of the recorded information. Thus, the term “video sequence” will be used to refer to any series of video frames, regardless of whether this series corresponds to a single camera shot (recorded between two cuts) or to a plurality of shots or scenes.
Traditionally, if a user desired to know what was the content of a particular video sequence he was obliged to watch as each frame, or a sub-sample of the frames, of the sequence was displayed successively in time. (For purposes of this document, the terms “he,” “him,” or “his” are used for convenience in place of she/he, her/him and hers/his, and are intended to be gender-neutral.) This approach is still wide-spread, and in applications where video data is accessed using a personal computer, the interface to the video often consists of a displayed window in which the video sequence is contained and a set of displayed controls similar to those found on a video tape recorder (allowing fastforward, rewind, etc.).
Developments in the fields of video indexing and video editing have provided other forms of interface to video information.
In the field of video indexing, it is necessary to code information contained in a video sequence in order to enable subsequent retrieval of the sequence from a database by reference to keywords or concepts. The coded content may, for example, identify the types of objects present in the video sequence, their properties/motion, the type of camera movements involved in the video sequence (pan, tracking shot, zoom, etc.), and other properties. A “summary” of the coded document may be prepared, consisting of certain representative frames taken from the sequence, together with text information or icons indicating how the sequence has been coded. The interface for interacting with the video database typically includes a computer input device enabling the user to specify objects or properties of interest and, in response to the query, the computer determines which video sequences in the database correspond to the input search terms and displays the appropriate “summaries”. The user then indicates whether or not a particular video sequence should be reproduced. Examples of products using this approach are described in the article “Advanced Imaging Product Survey: Photo, Document and Video” from the journal “Advanced Imaging”, October 1994, which document is incorporated herein by this reference.
In some video indexing schemes, the video sequence is divided up into shorter series of frames based upon the scene changes or the semantic content of the video information. A hierarchical structure may be defined. Index “summaries” may be produced for the different series of frames corresponding to nodes in the hierarchical structure. In such a case, at the time when a search is made, the “summary” corresponding to a complete video sequence may be retrieved for display to the user who is then allowed to request display of “summaries” relating to sub-sections of the video sequence which are lower down in the hierarchical structure. If the user so wishes, a selected sequence or sub-section is reproduced on the display monitor. Such a scheme is described in EP-A-0 555 028 which is incorporated herein by this reference.
A disadvantage of such traditional, indexing/searching interfaces to video sequences is that the dynamic quality of the video information is lost.
Another approach, derived from the field of video editing, consists of the “digital storyboard”. The video sequence is segmented into scenes and one or more representative frames from each scene is selected and displayed, usually accompanied by text information, side-by-side with representative frames from other segments. The user now has both a visual overview of all the scenes and a direct visual access to individual scenes. Each representative frame of the storyboard can be considered to be an icon. Selection of the icon via a pointing device (typically a mouse-controlled cursor) causes the associated video sequence or sub-sequence to be reproduced. Typical layouts for the storyboards are two-dimensional arrays or long one-dimensional strips. In the first case, the user scans the icons from the left to the right, line by line, whereas in the second case the user needs to move the strip across the screen.
Digital storyboards are typically created by a video editor who views the video sequence, segments the data into individual scenes and places each scene, with a descriptive comment, onto the storyboard. As is well-known from technical literature, many steps of this process can be automated. For example, different techniques for automatic detection of scene changes are discussed in the following documents, each of which is incorporated herein by reference:
Various methods for automatically detecting and tracking persons and objects in video sequences are considered in the following documents, each of which is incorporated herein by reference:
In the case of digital storyboards too, the dynamic quality of the video sequence is often lost or obscured. Some impression of the movement inherent in the video sequence can be preserved by selecting several frames to represent each scene, preferably frames which demonstrate the movement occurring in that scene. However, storyboardtype interfaces to video information remain awkward to use in view of the fact that multiple actions on the user's part are necessary in order to view and access data.
Attempts have been made to create a single visual image which represents both the content of individual views making up a video sequence and preserves the context, that is, the time-varying nature of the video image information.
One such approach creates a “trace” consisting of a single frame having superimposed images taken from different frames of the video sequence, these images being offset one from the other due to motion occurring between the different frames from which the images were taken. Thus, for example, in the case of a video sequence representing a sprinter running, the corresponding “trace” will include multiple probably overlapping) images of the sprinter, spaced in the direction in which the sprinter is running. Another approach of this kind generates a composite image, called a “salient still”, representative of the video sequence—see “Salient Video Stills: Content and Context Preserved” by Teodosio et al, Proc. ACM Multimedia 93, California, Aug. 1-6, 1993), pp 39-47 which article is incorporated herein by this reference in its entirety.
Still another approach of this general type consists in creation of a “video icon”, as described in the papers “Developing Power Tools for Video Indexinor and retrieval” by Zhang et al, SPIE, Vol.2185, pp 140-149-, and “Video Representation tools using a unified object and perspective based approach” by the present inventors, IS&T/SPIE Conference on Storage and Perusal for Image and Video Databases, San Jose, Calif., February 1995 which are incorporated herein by reference.
In a “video icon”, as illustrated in
Two special types of video icon have been proposed, “object based” video icons and video icons containing a representation of camera movement. In an “object based” video icon, as illustrated in
The video icons discussed above present the user with information concerning the content of the whole of a video sequence and serve as a selection tool allowing the user to access-frames of the video sequence out of the usual order. In other words, these icons allow non-sequential access to the video sequence. Nevertheless, the ways in which the user can interact with the video sequence information are strictly limited. The user can select frames for playback in a non-sequential way but he has little or no means of obtaining a deeper level of information concerning the video sequence as a whole, short of watching a playback of the whole sequence.
The present invention provides a novel type of interface to video information which allows the user to access information concerning a video sequence in a highly versatile manner. In particular, interactive video interfaces of the present invention enable a user to obtain deeper levels of information concerning an associated video sequence at positions in the sequence which are designated by the user as being of interest.
The present invention provides an interface to information concerning an associated video sequence, one such interface comprising:
information defining a three-dimensional root image, the root image consisting of a plurality of basic frames selected from said video sequence, and/or a plurality of portions of video frames corresponding to selected objects represented in the video sequence, x and y directions in the root image corresponding to x and y directions in the video frames and the z direction in the root image corresponding to the time axis whereby the basic frames are spaced apart from one another in the z direction of the root image by distances corresponding to the time separation between the respective video frames;
means for displaying views of the root image;
means for designating a viewing position relative to said root image; and
means for calculating image data representing said three-dimensional root image viewed from the designated viewing position, and for outputting said calculated image data to the displaying means.
According to the present invention, customized user interfaces may be created for video sequences. These interfaces comprise a displayable “root” image which directly represents the content and context of the image information in the video sequence and can be manipulated, either automatically or by the user, in order to display further image information, by designation of a viewing position with respect to the root image, the representation of the displayed image being changed in response to changes in the designated viewing position. In a preferred embodiment of the present invention, the representation of the displayed image changes dependent upon the designated viewing position as if the root image were a three-dimensional object. In such preferred embodiments, as the designated viewing position changes, the data necessary to form the displayed representation of the root image is calculated so as to provide the correct perspective view given the viewing angle, the distance separating the viewing position from the displayed quasi-object and whether the viewing position is above or below the displayed quasi-object.
In a reduced form, the present invention can provide non-interactive interfaces to video sequences, in which the root image information is packaged with an associated script defining a routine for automatically displaying a sequence of different views of the root image and performing a set of manipulations on the displayed image, no user manipulation being permitted. However, the full benefits of the invention are best seen in interactive interfaces where the viewing position of the root image is designated by the user, as follows. When the user first accesses the interface he is presented with a displayed image which represents the root image seen from a particular viewpoint (which may be a predetermined reference viewpoint). As he designates different viewing angles, the displayed image represents the root image seen from different perspectives. When the user designates viewing positions at greater or lesser distances from the root image, the displayed image increases or reduces the size and, preferably, resolution of the displayed information, accessing image data from additional video frames, if need be.
The customized, interactive interfaces provided by the present invention involve displayed images, representing the respective associated video sequences, which, in some ways, could be considered to be a navigable environment or a manipulable object. This environment or object is a quasi-three-dimensional entity. The x and y dimensions of the environment/object correspond to true spatial dimensions (corresponding to the x and y directions in the associated video frames) whereas the z dimension of the environment/object corresponds to the time axis. These interfaces could be considered to constitute a development of the “video icons” discussed above, now rendered interactive and manipulable by the user.
With the interfaces provided by the present invention, the user can select spatial and temporal information from a video sequence for access by designating a viewing position with respect to a video icon representing the video sequence. Arbitrarily chosen oblique “viewing directions” are possible whereby the user simultaneously accesses image information corresponding to portions of a number of different frames in the video sequence. As the user's viewing position relative to the video icon changes, the amount of a given frame which is visible to him, and the number and selection of frames which he can see, changes correspondingly.
As mentioned above, the interactive video interfaces of the present invention make use of a “root” image comprising a plurality of basic frames arranged to form a quasi-three dimensional object. It is preferred that the relative placement positions of the basic frames be arranged so as to indicate visually some underlying motion in the video sequence. Thus, for example, if the video sequence corresponds to a travelling shot moving down a hallway and tuning a corner, the envelope of the set of basic frames preferably does not have a parallelepiped shape but, instead, composes a “pipe” of rectangular section and bending, in a way corresponding to the camera travel during filming of the video sequence.
In preferred embodiments of video interfaces according to the present invention, the basic video frames making up the root image are chosen as a function of the amount of motion or change in the sequence. For example, in the case of a video sequence corresponding to a travelling shot, in which the background information changes, it is preferable that successive basic frames should include back-round information overlapping by, say, 50%.
In certain embodiments of the present invention, the root image corresponds to an “object-based video icon.” In other words, certain of the basic frames included in the root image are not included therein in full; only those portions corresponding to selected objects are included. Alternatively, or additionally, certain basic frames may be included in full in the root image but may include “hot objects,” that is, representations of objects selectable by the user. In response to selection of such “hot objects” by the user, the corresponding basic frames (and, if necessary, additional frames) are then displayed as if they had become transparent at all portions thereof except the portion(s) where the selected object or objects are displayed. The presence of such selectable objects in the root image allows the user to selectively isolate objects of interest in the video sequence and obtain at a glance a visual impression of the appearance and movement of the objects during the video sequence.
The interfaces of the present invention allow the user to select an arbitrary portion of the video sequence for playback. The user designates a portion of the video sequence which is of interest, by designating a corresponding portion of the displayed image forming part of the interface to the video sequence. This portion of the video sequence is than played back. The interface may include a displayed set of controls similar to those provided on a VCR in order to permit the user to select different modes for this playback, such as fast-forward, rewind, etc.
In preferred embodiments of interfaces according to the invention, the displayed image forming part of the interface remains visible whilst the designated portion of the sequence is being played back. This can be achieved in any number of ways, as for example, by providing a second display device upon which the playback takes place, or by designating a “playback window” on the display screen, this playback window being offset with respect to the screen area used by the interface, or by any other suitable means.
The preferred embodiments of interfaces according to the invention also permit the user to designate an object of interest and to select a playback mode in which only image information concerning that selected object is included in the playback. Furthermore, the user can select a single frame from the video sequence for display separately from the interactive displayed image generated by the interface.
In preferred embodiments, the interfaces of the present invention allow the user to generate a displayed image corresponding to a distortion of the root image. More especially, the displayed image can correspond to the root image subjected to an “accordion effect”, where the root image is “cracked open”, for example, by bending around a bend line so as to “fan out” video frames in the vicinity of the opening point, or is modified by linearly spreading apart video frames at a point of interest. The accordion effect can also be applied repetitively or otherwise in a nested fashion according to the present invention.
The present invention can provide user interfaces to “multi-threaded” video sequences, that is, video sequences consisting of numerous interrelated shorter segments such as are found, for example, in a video game where the user's choices change the scene which is displayed. Interfaces to such multi-threaded video sequences can include frames of the different video segments in the root image, such that the root image has a branching structure. Alternatively, some or all of the different threads may not be visible in the root image but may become visible as a result of user manipulation. For example, if the user expresses an interest in a particular region of the video sequence by designating a portion of a displayed root image using a pointing device (such as a mouse, or by touching a touch screen, etc.) then if multiple different threads of the sequence start from the designated area, image portions for these different threads may be added to the displayed image.
In preferred embodiments of interfaces according to the present invention, the root image for the video sequence concerned is associated with information defining how the corresponding displayed image will change in response to given types of user manipulation. Thus, for example, this associated information may define how many, or which additional frames are displayed when the user moves the viewing position closer up to the root image. Similarly, the associated information may identify which objects in the scene are “hot objects” and what image information will be displayed in relation to these hot objects when activated by the user.
Furthermore, different possibilities exist for delivering the components of the interface to the end user. In an application where video sequences are transmitted to a user over a telecommunications path, such as via the Internet, the user who is interested in a particular video sequence may first download only certain components of the associated interface. First of all he downloads information for generating a displayed view of the root image, together with an associated application program (if he does not already have an appropriate “interface player” loaded in his computer). The downloaded (or already-resident) application program includes basic routines for chancing the perspective of the displayed image in response to changes in the viewing position designated by the user. The application program is also adapted to consult any “associated information” (as mentioned above) which forms part of the interface and conditions the way in which the displayed image changes in response to certain predetermined user manipulations (such as “zoom-in” and “activate object”). If the interface does not contain any such “associated information” then the application program makes use of pre-set default parameters.
The root image corresponds to a particular set of basic video frames and information designating relative placement positions thereof. The root image information downloaded to the user may include just the data necessary to create a reference view of the root image or it may include the image data for the set of basic frames (in order to enable the changes in user viewing angle to be catered for without the need to download additional information). In a case where the user performs a manipulation which requires display of video information which is not present in the root image (e.g. he “zooms in” such that data from additional frames is required), this extra information can either be pre-packaged and supplied with the root image information or the extra information can be downloaded from the host website as and when it is needed.
Similar possibilities exist in the case of interfaces provided on CD-ROM. In general, the root image and other associated information will be provided on the CD-ROM in addition to the full video sequence. However, it is to be understood that, for reasons of space saving, catalogues of video sequences could be made consisting solely of interfaces, without the corresponding full video sequences.
In addition to providing the interfaces themselves, the present invention also provides apparatus for creation of interfaces according to the present invention. This may be dedicated hardware or, more preferably, a computer system programmed in accordance with specially designed computer programs.
Various of the steps involved in creation of a customized interface according to the present invention can be automated. Thus, for example, the selection of basic frames for inclusion in the “root image” of the interface can be made automatically according to one of a number of different algorithms, such as choosinbg one frame every n frames, or choosing 1 frame every time the camera movement has displaced the background by m%, etc. Similarly, the relative placement positions of the basic frames in the root image can be set automatically taking into account the time separation between those frames and, if desired, other factors such as camera motion. Similarly, the presence of objects or people in the video sequence can be detected automatically according to one of the known algorithms (such as those discussed in the references cited above), and an “object oriented” root image can be created automatically. Thus, in some embodiments, the interface creation apparatus of the present invention has the capability of automatically processing video sequence information in order to produce a root image. These embodiments include means for associating with the root image a standard set of routines for changing the representation of the displayed image in response to user manipulations.
However, it is often preferable actively to design the characteristics of interactive interfaces according to the invention, such that the ways in which the end user can interact with the video information are limited or channeled in preferred directions, This is particularly true in the case of video sequences which are advertisements or are used in educational software and the like.
Thus, the present invention provides a toolkit for use in creation of customized interfaces. In preferred embodiments, the toolkit enables a designer to tailor the configuration and content of the root image, as well as to specify which objects in the video sequence are “hot objects” and to control the way in which the displayed interface image will change in response to manipulation by an end user. Thus, among other things, the toolkit enables the interface designer to determine which frames of the video sequence should be used as basic frames in the root image, and how many additional frames are added to the displayed image when the user designates a viewing position close to the root image.
According to another aspect, the invention relates to network ditribution and management of interactive video and multi-media containers. A need exists for methods and systems for transmitting video and other multi-media files across a network, such as the Internet. U.S. Pat. No. 5,956,716 to Kenner et al. provides an example of a system and method for the delivery of video data over a computer network. In Kenner, a user uses a multimedia terminal to send a request for video clips from a database. A local storage and retrieval module receives and processes video clip requsts and a primary index manager causes the distribution of video clips among a plurality of extended storage and retrieval modules. The extended storage and retrieval modules store a plurality of databases including those that contain video clips. A data sequencing interface directs the extended storage and retrieval module to download the requested video clips. The video clips are then downloaded to the multimedia terminal via the local storage and retrieval module.
Systems and methods according to the invention provide for the network distribution and management of interactive video and multi-media containers. Systems and methods not only can distribute video and other multi-media files but they can also distribute multi-media containers. Consequently, users would be able to access information concerning the mult-imedia files in a highly versatile manner. Systems and methods according to the invention also enable for the transmission of information both to and from the users. Thus, systems and methods according to the invention provide for colloboration between users. For instance, work performed by one user in indexing or in providing annotations is not restricted to just that user but can be shared with others having access to the multi-media file. Other advantages and benefits of the invention are provided in the following description and will be apparent to those skilled in the art.
Further features and advantages of the present invention will become apparent from the following description of preferred embodiments thereof, given by way of example, and illustrated by the accompanying drawings, in which:
I. Interactive Interface
The components of an interactive interface according to a first preferred embodiment of the present invention will now be described with reference to FIG. 2. In this example, an interactive interface of the invention is associated with video sequences recorded on a CD-ROM.
As shown in
According to the first embodiment of the invention, the CD-ROM has recorded thereon not only the video sequence image information 8 (in any convenient format), but also a respective interface data file (FDIi) 10 for each video sequence, together with a video interface application program 11. The content of a typical data file is illustrated in FIG. 3. Respective scripts 12 are optionally associated with the interface data files. When data on the CD-ROM is to be read, the video interface application program 11 is operated by the central processor portion 2 of the computer system and the interface data file applicable to the video sequence selected by the user is processed in order to cause an interactive video icon (see, for example,
The types of manipulations of the interactive video icon which are available to the user will now be described with reference to
When the user seeks to explore the video sequence via the interactive video icon displayed on the display screen, one of the basic operations he can perform is to designate a position on the screen as a viewing position relative to the displayed image (e.g. by “clicking” with the computer mouse). In
The above-mentioned cuboid is a special case of a “root image” according to the present invention. This “root image” is derived from the video sequence and conveys information concerning both the image content of the selected sub-set of frames (called below, “basic frames”) and the relative “position” of that image information in time as well as space. It is to be appreciated that the “root image” is defined by information in the interface data file. The definition specifies which video frames are “basic frames” (for example, by storing the relevant frame numbers), as well as specifying the placement positions of the basic frames relative to one another within the root image.
The central processor portion 2 of the computer system calculates the image data required to generate the displayed image from the root image definition contained in the appropriate interface data file, image data of the basic frames (and, where required, additional frames) and the viewing position designated by the user, using, standard ray-tracing techniques. The data required to generated the displayed image is loaded into the video buffer and displayed on the display screen.
According to the present invention it is preferred that, when the user designates a viewing position close up to the interactive video icon, the image information in the area of interest should be enriched. This is achieved by including, in the displayed image, image data relating to additional video frames besides the basic video frames. Such a case is illustrated in
Preferably, the interface data file includes data specifying how the choice should be made of additional frames to be added as the user “moves close up” to the displayed image. More preferably, this data defines rules governing the choice of how many, and which, additional frames should be used to enrich the displayed image as the designated viewing position changes. These rules can, for example, define a mathematical relationship between the number of displayed frames and the distance separating the designated viewing position and the displayed quasi-object. In preferred embodiments of the invention, the number of frames which are added to the display as the viewing position approaches the displayed quasi-object depends upon the amount of motion or change in the video sequence at that location.
The example illustrated in
For example, the relative placement positions of the basic frames may be selected such that the envelope of the root image has a shape which reflects motion in the corresponding video sequence (either camera motion, during tracking shots and the like, or motion of objects represented in the sequence)—see the corresponding interactive icon shown in FIG. 6A. Similarly, the dimensions of the basic frames in the root image may be scaled so as to visually represent a zoom effect occurring in the video sequence -see the corresponding interactive icon shown in FIG. 6B.
It will be seen that the interactive icon represented in
If the user expresses an interest in either of the two objects, for example, by designating a screen position corresponding to one of the objects (e.g. by “clicking” on the left-hand person using the right-hand mouse button), then the interface application program controls the displayed image such that extraneous portions of the displayed frames disappear from the display, leaving only a representation of the two people and their motion, as shown in FIG. 7B. Thus, the objects of interest are “extracted” from their surroundings. The “missing” or transparent portions of the displayed frames can be restored to the displayed image at the user's demand (e.g. by a further “click” of the mouse button).
It is to be understood that, according to the present invention, interfaces may be designed such that particular “extractable” objects may be extracted simultaneously with some or all of the other extractable objects, or they may be extracted individually. Sophisticated interfaces according to the present invention can incorporate object-extraction routines permitting the user to arbitrarily select objects visible in the displayed view of the root image, for extraction. Thus, for example, the user may use a pointing device to create a frame around an object visible in a displayed view of the root image and the application program then provides analysis routines permitting identification of the designated object in the other basic frames of the root image (and, if required, in additional frames) so as to cause display of that selected object as if it were located on transparent frames.
It may be desirable to allow the user to obtain a close-up view of a particular portion of the interactive video icon in a manner which does not correspond to a strict perspective view of the reion concerned. Preferred embodiments of interface according to the invention thus provide a so-called “accordion” effect, as illustrated in FIG. 8. When the user manipulates the icon by an “accordion” effect at a particular point, the basic frames in the vicinity of the region of interest are spread so as to provide the user with a better view. Further, preferably, the function of displaying additional frames so as to increase detail is inhibited during the “accordion” effect.
In the case of “multi-threaded” video sequences, such as are traditionally found in video-based computer games and educational software and involve parallel video subsequences which are accessed alternatively depending upon the user's choices, these too can be the subject of interfaces according to the present invention. In such a case, the interface designer may choose to include frames from different parallel video subsequences in the interface's root image in order to give the user an idea of the different plot strands available to him in the video sequence.
Alternatively, or additionally, the designer may create secondary root images for the respective sub-sequences, these secondary root images being used to generate the displayed image only when the user designates a viewing position close to the video frame where the sub-sequence begins. In the case of interfaces to such computer games or educational software, this is a logical choice since it is at the point where the video sub-sequence branches from the main sequence that user choices during playing of the game, or using of the educational software, change the experienced scenario.
Another manipulation which it is preferable to include in interfaces according to the invention is the traditional set of displayed VCR controls which permit the user to playback the video sequence with which the displayed video icon is associated. Furthermore, the user can select for playback portions or frames within the sequence by, for example, “clicking” with the mouse button on the frames of interest as displayed in the interactive video icon. The video playback can take place on a separate display screen or on a window defined on the display screen displaying the video icon.
As mentioned above, a particular video sequence may be associated with an interface data file and a script. The script is a routine defined by the interface designer which leads the user through the use of the interface. The script can, for example, consist of a routine to cause an automatic demonstration of the different manipulations possible of the displayed quasi-object. The user can alter the running of the script in the usual way, for example by pausing it, slowing it down, etc.
The script may, if desired, include additional text, sound or graphic information which can be reproduced in association with the displayed view of the root image either automatically or in response to operations performed by the end user. Script functionality according to the present invention allows creation and editing of viewing scenarios that may be subsequently be played, in part or in whole, automatically, or interactively with user inputs. For example, in a completely automatic mode, the user can cause the scenario to begin to play by itself and take the user through the scenario and any associated information by simply reading the scenario and changing the view. In other situations the script may call for interaction by the user, such as to initiate a transaction. In this case the user may be asked to specify information, e.g. if he wants to purchase the video or any other items associated with what has been viewed. In yet other situations the editor may leave visible tags which when activated by the user will cause some information to be displayed on the display device; e.g. associated text, graphics, video, or sound files which are played through the speakers of the display device. In certain cases these tags are attached to objects selected and extracted from the video sequence, such as so-called “hot objects” according to the present invention.
The present invention provides toolkitd for use by designers wishing to create an interactive video interface according to the present invention. These toolkits are preferably implemented as a computer program for running on a general purpose computer. The toolkits present the designer with displayed menus and instructions to lead him through a process including steps such as the typical sequence illustrated in FIG. 10.
The designer first of all indicates for which video sequence he desires to create an interface, for example by typing in the name of a stored file containing the video sequence information. Preferably, the toolkit accesses this video sequence information for display in a window on the screen for consultation by the designer during the interface design process. In such preferred embodiments of the toolkit, the designer may make his selection of basic frames/objects for the root image, extractable objects and the like by stepping slowly through the video sequence and, for example, using a mouse to place a cursor on frames or portions of frames which are of interest. The toolkit logs the frame number (and x, y locations of regions in a frame, where appropriate) of the frames/frame portions indicated by the designer and associates this positional information with the appropriate parameter being defined. Preferably, at the end of the interface design process the designer is presented with a displayed view of the root image for manipulation so that he may determine whether any changes to the interface data file are required.
Different versions of the application program can be associated with the interface data file (and script, if present) depending upon the interface functions which are to be supported. Thus, if no script is associated with the interface data file, the application program does not require routines handling the running of scripts. Similarly, if the interface data file does not permit an accordion effect to be performed by the end user then the application program does not need to include routines required for calculating display information for such effects. If the interface designer believes that the end user is likely already to have an application program suitable for running interfaces according to the present invention then he may choose not to package an application program with the interface data file or else to associate with the interface data file merely information which identifies a suitable version of application program for running this particular interface.
The present invention has been described above in connection with video sequences stored on CD-ROM. It is to be understood that the present invention can be realized in numerous other applications. The content of the interface data file and the elements of the interface which are present at the same location as the end user can vary depending upon the application.
For example, in an application where a video sequence is provided at a web-site, the user may first download via his telecommunications connection just the interface data file applicable to the sequence. If the user does not already have software suitable for handling manipulation of the interactive video icon then he will also download the corresponding application program. As the user manipulates the interactive video icon, any extra image information that he may require which has not already been downloaded can be downloaded in a dynamic fashion as required.
This process can be audited according to the present invention if desired. The user's interaction with the interface can be audited, and he can interact with the transaction/audit functionality for example to supply any information required by a script which may then be recorded and stored. Depending upon the application, the transaction/audit information can be stored and made available for externally (optional) located auditing and transaction processing facilities/applications. In a typical situation, the auditing information can be transmitted at the end of a session whereas the transaction information may be performed on-line, i.e. the transaction information is submitted during the session. Real time transmission can also occur according to the present invention, however.
Another example is the case of a catalogue on CD-ROM including only interfaces rather than the associated video sequences, in order to save space. In such a case, rather than including a pointer to the image information of the basic frames of the root image, the interface data frame includes the image information. Some additional image information may also be provided.
The following disclosure relates to a preferred implementation according to the present invention, with reference to
A. Interface Editor Unit
Editors, readers and viewers according to the present invention can be implemented in hardware, hardware/software hybrid, or as software on a dedicated platform, a workstation, a personal computer, or any other hardware. Different units implemented in software run on a CPU or graphics boards or other conventional hardware in a conventional manner, and the various storage devices can be general purpose computer storage devices such as magnetic disks, CD-ROMs, DVD, etc.
With reference to
The creation of an interface using the editor is discussed below in three phases: (1) Analysis, (2) Visual layout and (3) Effects creation.
The video document chosen by the editor is first processed by the activity measure unit (103). The activity measure unit is responsible for computing various parameters related to the motion and changes in the video. This unit typically will implement one of a number of known techniques for measuring changes, e.g., by calculating the statistics of the differences between frames, by tracking objects in motion, or by estimating camera motions by separating foreground and background portions of the image. In other implementations this unit may use motion vector information stored in an MPEG-encoded sequence to detect important frames of activity in the video document. The activity measures template store is optional but would contain templates which can be used to calculate the frame ranking measure and could be specified by the user through the user interaction unit.
These parameters are then used to calculate a frame ranking measure which ranks the different frames as to whether they should be included in the interface. The frame ranking measure is derived heuristically from these measures [e.g., by normalizing the values and taking an average of the parameters, and can be tailored for different kinds of sequences (traveling shots, single objects in motion, etc) or applications]. The editor may choose a pre-defined set of parameters from the activity measures template store (108) to detect or highlight a specific kind of activity (rapid motion, abrupt changes, accelerations, etc.)
The frame ranking measures can be employed by the user acting through the user interaction unit on the frame selection unit (104) to select the frames to be included within the interface. For example, if 10 frames are to be included in the interface then in default mode the 10 frames corresponding to the 10 largest frame making measures are selected for inclusion in the interface. The user can then interactively de-select some of these frames and add other frames.
The camera motion analysis unit (105) is an optional unit which typically will implement one of a number of known techniques for measuring camera motion parameters. This information can be used to determine what shape to give to the outer envelope of the interface as shown in
The object selection unit (106A) is responsible for selecting or detecting individual objects in the video document. There are various modes possible: in a completely manual mode the editor may visually select and outline an object of interest in a given frame through the user interaction unit (120); in a semi-manual mode, the editor simply points at an object and chooses from the object templates store (107) features and associated algorithms to use for extracting and tracking the chosen object; in another mode the editor may chose one of a set of pre-defined templates of objects and known pattern matching techniques are used to detect whether any objects of interest are preset. The user may even assign a name/identifier to the object and add the object to the object templates store (107). In this latter case searches for multiple occurrences of the same object can be initiated by the user. The information regarding the properties of the object may be optionally stored in the FDI file.
The object extraction and tracking unit (106B) is now responsible for extracting the object of interest from the frame and then tracking it by using known tracking algorithms. The algorithms used are either chosen by the user or by default. It is understood that the object selecting, detection, extraction, and tracking process may be highly interactive and that the user may be called upon or choose to intervene in the process a number of times. The information about the presence and location of objects may be optionally stored in the FDI file.
In certain applications the FDI file can be made available to an external program, for example when the interface editor is associated with an indexing program, the task of which is to attach indexes (identifiers) to the video documents, to portions thereof, or to objects located within the video document.
2. Visual Layout.
The user acting through the user interaction unit (120) on the interface creation unit (109) determines the visual layout of the interface.
He can shape the outer envelope of the interface in any way that he desires; two examples are provided in
The different pieces of information generated by the units described above are gathered together by the interface creation unit (109) into an FDI file containing a description of the interface in terms of its layout i.e. shape and structure, the image frame numbers and their positions, and if available, the extracted features the ranking of the frames and the camera motion information. This information is transmitted to the interface effects creation unit (117).
3. Effects Creation.
The editor can also specify three classes of interface features which serve to convey additional information to the user and which allow the user to interact with the interface. The editor performs this specification through the interface effects creation unit (117).
The zooming effects creation unit (110) is used by the editor to specify which frames will be made visible, and also which will be rendered invisible to the user when he moves up closer to the interface (
The special effects creation unit (111) is used by the editor to create special visual effects on the interface. One such example is the accordion effect illustrated in
The script effects creation unit (113) allows the editor of the interface to build an interface viewing scenario that may be subsequently be played, in part or in whole, automatically, or interactively with user inputs. For example, in a completely automatic mode when the user calls up the interface it begins to play by itself and takes the user through the interface and any associated information by simply reading the scenario and changing the view of the interface. In other situations the script may call for the user to interact with the interface, e.g. to initiate a transaction. In this case the user may be asked to specify information, e.g. if he wants to purchase the video or any other items associated with the interface. In yet other situations the editor may leave visible tags which when activated by the user will cause some information to be displayed on the display device; e.g. associated text, graphics, video, or sound files which are played through the speakers of the display device. In certain cases these tags are attached to objects selected and extracted from the video sequence by units 6A and 6B and become so-called “hot object.” The editor creates the scripts by calling up templates from the script effects templates store (115) and instantiating them by defining the tag and the locations of the information to be called up.
The interface effects creation unit (117) creates 4 files which are passed to the interface database manager (118) which will store these files either remotely or locally as the case may be: (1) The FDI file, completed by the special effect and script tags, text and graphics which have been added to the interface and which are directly visible to the user. (2) The zoom effect details, scripts and special effects. (3) The application programs (optional) to view the interface; i.e., allow the user to view the interface from different perspectives, traverse the interface, run the script, perform the special effects, or coded information which indicate which application program residing on the users machine can be used to perform these operations. (4) The video sequence and any other associated information (data) required for reading the interface.
These files are shown stored in storage unit (119) but depending upon the embodiment they may be physically located in the same storage device, in separate storage devices (as shown) either locally (as shown) or remotely.
During the editing process, the user/editor can view the interface under construction, according to the current set of parameters, templates and designer preferences, on the interface viewer unit (121) (presented in FIG. 12 and described below), thus allowing the editor to interactively change its appearance and features.
B. Interface Viewer Unit
Having chosen an interface through a traditional method, for example by using a database query language or by using a browser such as are used for viewing data on the Web, the interface viewer unit is then employed to read and interact with the interface.
In a typical application the storage units (201) are remotely located and accessed through the interface database manager (202) by way of a communication channel or network; depending upon the size and characteristics of the channel and the application the interface data may be loaded in its entirety or fetched on a as need basis.
The data are then stored in a local memory unit (203) which may be either a cache memory, a disk store or any other writable storage element. The local memory unit (203) stores the 4 files created by the editor (see above) and in addition a transaction/audit file. In certain cases the applications programs are already resident in the interface viewer unit and so do not need to be transmitted.
The CPU unit (204) fetches the application program, deduces which actions need to be performed, and then fetches the relevant interface information contained in the local memory unit (203). Typically the CPU unit fetches the required application program for the user interaction unit (205), the navigation unit (206), and the transaction/audit unit (207), then interface information is read from the local memory unit (203) passed to the interface renderer unit (208) which then calculates how the interface is to appear or be rendered for viewing on the display device (209).
The user interacts with the interface through the user interaction unit (205) to the navigation unit (206) and all his actions are audited by the transaction/audit unit (207). In addition, the user can interact with the transaction/audit unit (207) for example to supply any information required by the script which is then recorded and stored in the transaction/audit portion of the local memory unit (203). Depending upon the application, this transaction/audit file or a portion thereof is transmitted by the interface database manager to the appropriate storage unit (201). This information is then available for externally (optional) located auditing and transaction processing facilities/applications. In a typical situation, the auditing information is transmitted at the end of the session whereas the transaction information may be performed on-line, i.e. the transaction information is submitted during the session.
Through the navigation unit (206) the user can choose the point of view from which to view the interface (or a portion of the interface). The interface rendered unit (208) then calculates how the interface is to appear or be rendered for viewing on the display device (209).
If the user chooses to zoom in or zoom out, then the zoom effects unit (210) fetches the required application program, reads the zoom effect parameters stored in the local memory store (203), determines the frames to be dropped or added and supplies this information (including the additional frames if needed) to interface renderer unit (208) which then calculates how the interface is to appear or be rendered for viewing on the display device (209).
If the user chooses to view part of the underlying video then the video play effects unit (211), fetches the required application program, then reads the required video data from the local memory unit (203) and plays the video on a second display device (209) or in a new window if only one display device is available.
If the user chooses to interact with a hot pre-extracted object (created by the special effects unit), then the special effects unit (212), fetches the required application program, reads the locations of the object and the corresponding frames are modified so as to be transparent wherever the objects do not occur; the new frames are passed to interface renderer unit (208) which then calculates how the interface is to appear or be rendered for viewing on the display device (209). In cases where the extracted object is to be played as a video the frames are passed to the video effects unit (211) which then plays the video on a second display device (209) or in a new window if only one display device is available. Similarly if the user chooses to view an accordion effect then the special effects unit fetches the accordion effect store (203), determines the frames to be dropped or added and calculates parameters stored in the local memory the relative position of all the frames and supplies this information (including the additional frames if needed) to interface renderer unit (208) which then calculates how the interface is to appear or be rendered for viewing on the display device (209).
If the user designates a tag created by the script then the script effects unit (214) fetches the required application program, reads the corresponding portion of the script and the related information required to carry out the portion of the script associated with the tag designated. If the interface is to be played in automatic mode then the script effects unit (214) fetches the entire script and all the related information required to carry out the script. When needed the zoom effects unit (210), the video play unit (211), and the special effects unit (212) may be called into play. If the script calls for user input such as required for carrying out a transaction, then a new window may be opened on the display device (or on a second display device) where the information is supplied and transmitted to the transaction/audit unit (207). In semi-automatic mode control of the viewing of the interface is passed between the script effects unit (214) and the navigation as instructed by the user through the user interaction unit (205).
Although the above-discussed preferred embodiments of the present invention present certain combinations of features, it is to be understood that the present invention is not limited to the details of these particular examples. Firstly, since image processing is performed on image data in digital form, it is to be understood that in the case where the video sequence consists of data in analogue form, an analogue-to digital converter or the like will be used in order to provide image data in a form suitable for processing. It is to be understood that the present invention can be used to create interfaces to video sequences where the video data is in compressed form, encrypted, etc. Secondly, references above to user input or user selection processes cover the use of any input device whatsoever operable by the user including, but not limited to, a keyboard, a mouse (or other pointing, device), a touch screen or panel, glove input devices, detectors of eye movements, voice actuated devices, etc. Thirdly, references above to “displays” cover the use of numerous different devices such as, but not limited to, conventional monitor screens, liquid crystal displays, etc.
Furthermore, for ease of comprehension the above discussion describes interfaces according to the present invention in which the respective root images each have a single characteristic feature, such as, giving a visual representation of motion, or giving a visual representation of zoom, or having a multi-threaded structure, etc. It is to be understood that a single root image can combine several of these features, as desired. Similarly, special effects such as object extraction, the accordion effect, etc. have been described separately. Again, it is to be understood that interfaces according to the invention can be designed to permit any desired combination of special effects.
II. Global Topology
1 Glossary and Concepts
Obvious Database Annotation
Obvious Streaming Format
Synonym for Obvious Site
The Obvious Site Manager is the main server in a site. It constitute the entry- point of a site and manages all others servers.
Obvious Administration Server.
Obvious Asset Manager.
Obvious Site Directory.
Obvious Media Server.
The Obvious Network Architecture is a distributed systems composed of several software components. These components can be grouped into 5 categories:
1. Runtime Components
The Runtime Components constitute the core elements that are involved during the visualisation of OBVIs. When an OBVI is opened from a client application, several Runtime Components are used for streaming the video, retrieving the images of the blocks, accessing the annotations and the structure, etc.
2. Video Components
The Video Components are involved in the process of video acquisition, storage, registering of media pieces. Video Components play an important role in the Obvious Network Architecture. OBVIs are bound to a video and, except for OVI files created from scratch, this video must be properly registered in the system before it can be used.
3. OBVI Components
The OBVI Components concern the authoring, indexing and publishing of OBVIs. OBVIs are created from a registered or a non-registered video file. Then, they are published on the system as database objects. Their content is indexed and end-users can do a search for retrieving an OBVI in various forms.
4. Distribution Components
The Distribution Components are involved for the distribution of OBVIs from a repository location to the end-user. Some of these components simply use traditional distribution channels (Email, Web, FTP). Others try to embed OBVIs into video streams (ASF, RM, QT).
5. Database Components
The Indexing Components offer indexing facilities for the others components. These components use database technologies for providing a reliable and efficient way to store, index, query and transact on various objects in the system. In particular, Database Components handle video, OBVIs, streams, users, machines as database objects.
2.1 Runtime Components
The Runtime Components are detailed in FIG. 14. This diagram shows that Runtime Components interact directly with client applications and with the Database Components. Runtime Components include several types of servers: video server, web server, OSM, OMS, OSD.
2.1.1 Obvious Site Directory
The Obvious Site Directory is a directory service for Obvious Sites. By using the services of an Obvious Site Directory, an end-user can get the list of all available Obvious Sites in the world with their characteristics, security policies, description of their content, etc. All sites that are Obvious-compliant, i.e. being using our technology or being hosted on our portal, will have an entry in this directory.
An OBVI contains a Site Identifier. A Site Identifier is a number that uniquely references a site. The mapping between the Site Identifier and the actual address of the site involves a mechanism called Site Resolution. This mechanism is described in below in Section V entitled The Obvious Site Directory.
If the administrator of a site does not want to be referenced on this directory then the site is said to be autonomous. In that case, site Resolution is done locally, at each client application, since only one site/IP pair is necessary.
2.1.2 Obvious Site Manager
This component is responsible for the management of an Obvious Site. It handles access control and service replication. Client applications must first connect to an Obvious Site Manager before accessing resources in a given Obvious Site. The Obvious Site manager is the entry point for a given portal.
2.1.3 Obvious Media Server
The Obvious Media Server is involved during the visualisation process of an OBVI. Its primary goal is to distributes still images corresponding to OBVI blocks (because video servers can not distribute still images efficiently). The OMS allows remote applications to retrieve individual still images that can be used for 2D/3D storyboard, thumbnails of video documents, etc. Aside the distribution of still image, the OMS also accomplished important tasks such as the distribution of annotations and structure, the generation of OVI and XML files, etc.
2.1.4 Video Server
Several video servers can be used in the Runtime Components. The architecture does not rely on a specific streaming technology. Any third-party video server can be incorporated in the Obvious Network Architecture, if client applications (such as the Obvious Media Viewer) can support it. For example, the OMM/OME/OMV suite of applications support RealServer G2, NetShow and NetShow Theater streams.
RealServer G2 and NetShow should be used on low-bandwidth networks. NetShow Theater is the preferred choice for delivering broadcast-quality MPEG video streams across high-bandwidth networks and intranets.
2.1.5 Web Server
In the Obvious Network Architecture, several web servers can be used for accessing annotations. These servers distribute any content type that can be displayed in the IE-based annotation viewer of the OMM/OME/OMV tools. When an OVI file is published/indexed, its embedded annotations are converted into HTML and automatically published on these Web servers.
2.1.6 External Database
In the case of database annotations (ODA), several external databases can be used. Access to these databases is handled by a specific plugin, via ODBC, ADO or any other database access mechanism.
2.2 Video Components
The Video Components are detailed FIG. 15. These components interact with the Database Components and the OBVI Components.
2.2.1 Video Cataloging Tools
Video cataloging tools perform the following tasks:
In current implementation, these tasks are accomplished from a unified interface: the Obvious Management Console. The Obvious Management Console, described in details below in section XI, entitled Obvious Management Console, is an application that allows the administration of the whole system. In particular, it allows to run these video cataloging tools.
Video analysis and video archiving are handled by the Obvious Asset Manager.
2.2.2 Obvious Asset Manager
The Obvious Asset manager has 2 roles. It acts as a repository for the video files, for archiving purposes and it also hosts the VAMT for video analysis purposes.
Concerning video archiving, the Obvious Asset Manager is basically an FTP server that allows remote clients applications to upload their video files. This upload is never done manually. It is automatically handled from the Obvious Management Console, when the user registers a new video file. Video registering (composed of Video Document registering and Media registering) plays an important role in the Obvious Network Architecture.
Concerning video analysis, 3 modules are used: the VAMT Engine, the VAMT Service and the VAMT Manager. The VAMT Engine is a server that runs several simultaneous analysis jobs. It internally uses the Obvious VAMT Engine which handles the core analysis. Each analysis job involves the analysis of a specific video file from a timecode in to a timecode out. The Obvious VAMT Manager is the graphical user interface that allows remote administration of the Obvious VAMT Service. The Obvious VAMT Manager allows the user to define, launch, stop and edit analysis jobs. These modules are described in details in FIG. 15 and in more detail in Section IX entitled the Obvious Asset Server.
2.3 OBVI Components
2.3.1 Authoring and Editing Applications
There is only one authoring tool in the current state of development: the Obvious Media Manager. The Obvious Media Editor is an editing tool.
2.3.2 Viewer Applications
The Obvious Media Viewer is a viewer application implemented in Visual Basic and Visual C++. The Obvious Java Viewer is a cross-platform, lightweight and web-enabled OBVI viewer.
2.3.3 Obvious Publisher
The Obvious Publisher is the end-user graphical interface that allows to publish and index OVI files. Publishing involves the transformation of the OVI file into an OBVI object that can be stored in the OIS database. Indexing concerns the OBVI metadata and annotations. In current version, the indexing process can use the Microsoft Index Server or the Oracle 8 database (with ConText Cartridge). In both cases a specific filter is needed to parse the OVI files, extract meaningful information and populate index information. The Obvious Publisher doesn't do the publishing and the indexing job. It is a GUI that collects all the information necessary for publishing. The OVI file and the collected information are then uploaded to another machine where resides the Obvious Publishing Engine.
2.3.4 Obvious Publishing Engine
The Obvious Publishing Engine is the core module that handles the publishing and indexing of an OVI file. It acts as a daemon that automatically takes OVI files, parse them, publish their annotations, and create corresponding entries in the database.
2.3.5 Obvious Publishing Manager
The Obvious Publishing Manager is an application that can be used by administrators to remotely control and manager the Obvious Publishing Engine. From that application, the user can see the status of the publishing/indexing process, see the number of jobs, configure the scanned directories, etc.
2.4 Distribution Components
2.4.1 Classical Distribution
OVI files can be distributed by Web servers, FTP servers or by email. An OVI can be bound to its meia content in several ways. Concerning video, it can have the whole media file embedded or it can be linked to a registered Vdoc/Media. Concerning the annotations, they can be embedded or distributed on-demand at runtime.
2.4.2 On Demand Distribution
The Obvious Media Server is able to distribute OBVIs on demand (as OVI, XML or OSF files) or multicast (as OSF files). The distributed OVI, XML and OSF files can be either generated on the fly or pre-calculated.
2.4.3 Streamed Distribution
The Obvious Streaming Architecture defines the mechanisms that allows:
These elements are detailed in section XVI, entitled The Whole Picture.
2.4.4 Immerse Streamed Distribution
Several hardware/software technologies allow the encapsulation of bit-streams, byte-streams, and IP data into MPEG-2 packets. These MPEG-2 packets can then be injected into a DVB-compliant transport stream that can be carried over satellite, cable, or terrestrial digital transmission systems. By combining audio/video and OBVI data streams, these technologies enable a multitude of point-to-multipoint OBVI delivery applications for traditional broadcasters looking for new business models.
QuickTime movies can handle user-defined tracks of information. A new track can be created for carrying OBVI data in a traditional QuickTime movie. The track identifier, as defined by the QuickTime specification, is “OBVI”. The embedding of an OBVI in a QuickTime is an alpha functionality. It is available in the OMM and is implemented in a separate DLL called LibQT.dll. It currently uses QuickTime 3. Porting to QuickTime 4 is expected in next major release of the OMM.
A new QuickTime player has also been implemented. It acts as a classic QuickTime player. However, if the input movie file contains the OBVI track, a dialog box asks the user if he wants to launch the OMM. In that case, the OVI file is extracted from the QuickTime file is opened with the OMM.
Microsoft has recently defined a new ASF format. NetShow 2.0 was using ASF version 1.ASF version 2, available in NetShow 3.0 is more powerful and flexible. It allows the embedding of user-defined packets of data. Experiments have shown that embedding OBVI data into an ASF stream should be easy. However, we should wait until the end of the OSF specification because it should be possible to directly embed OSF packets in an ASF stream.
The RealServer G2 SDK exposes a way for embedding user-defined packets of data in a RealMedia stream. Here again, there will be a strong relation with OSF. OSF packets can be directly embedded in RealMedia streams.
2.5 Database Component
The Database Component is also called the Obvious Indexing System (OIS). It acts as a repository for registering and indexing the various entities in the system: video documents, media, streams, OBVIs, users, groups, etc. The Database Component interact with all others components.
3 Implementation Issues
3.1 Technologies Involved
The Obvious Network Architecture is designed for Windows technologies. Microsoft Internet Information Server is extensively used in servers that need to provide services to remote application clients using HTTP. Others servers. A three-tier model is used. The OIS constitute the backend store, the database. The business logic, as defined in the three-tier model, is implemented as a set of ISAPI scripts and NT services. Client applications interact with the system by using HTTP. XML is heavily used at different levels as a standard format for data exchange between client/server and server/server communications.
Some parts of the system use direct TCP/IP communications, essentially for internal server/server data exchange.
All server components are implemented in Visual C++/MFC. This include NT services (implemented with ATL) and ISAPI extensions. MFC is used for all graphical parts.
Server components can easily be ported to others platform, especially Unix. ISAPI scripts can directly be converted into CGI scripts. NT services can be rewritten as Unix daemons. Windows-specific code has been, wherever possible, coded in a separate DLL.
Concerning the database, standard built-in types have been used in the definition of the tables. Data access is handled by ADO and should be easily replaced buy direct ODBC calls, if necessary.
III. THE OBVI
1 The OBVI
An OBVI is not a file. An OBVI is a hyper media container. This general definition does not imply any particular storage format. However, in the current implementation of the Obvious Architecture, an OBVI is stored and managed as a database object. This is the primary storage format.
From this primary, database-centric, storage format, an OBVI can the be exported in any additional storage format (an XML file, an OVI file or an OSF file), called secondary storage formats. In particular, the OVI file format is one of the storage format that has been defined fort storing OBVI objects outside the database, on a regular file-system. The OVI file format is a way to store an OBVI in a binary file. This file can be read by client applications for visualising and interacting with the OBVI.
Historically, the first OBVI format storage that has been developed is the OVI file format. This is the format that is natively read from the OMM/OME/OMV suite of client applications. However, this file-based storage format is only one of the possible secondary storage formats to which an OBVI can be exported.
The Obvious Network Architecture, described in details in this document, focuses on the specification of the primary storage format: the database format. It will also show how OVI files can be converted into OBVIs (this is the publishing/indexing process) and how OBVIs can be extracted from the database in an OVI or a XML form.
In current version of the OVI file format, the following annotations types can be found:
An annotation can be either embedded in the OVI file or located on another server. In the second case, the OVI stores the URL of the remote annotation. While the internal mechanism is already implemented in the OBVIKernel, HTML annotations are the only kind of annotations that fully use this feature: the user can create embedded HTML annotation or linked HTML annotations.
However, as it will explained in future chapters, the Obvious Network Architecture heavily relies on this mechanism for providing a global framework for publishing indexing and distributing OBVIs. Future versions of the OMM will implement this dual mechanism for all others annotations types.
2.1 HTML Annotation
In an OVI file, HTML annotations are internally represented by a description file. This is a text file describing the material that constitute the HTML annotations. In the case of a remote HTML annotation, the description file contains the full URL of the document. In the case of an embedded HTML annotation, the description file gives the names of all temporary files that constitute the annotation. These can be HTML files, JAVA applets, images or any other file that can be part of an HTML page.
To ensure a proper migration into the Obvious Network Technology framework, this description file will be encoded in XML. This change is expected for next major release of the OMM.
Another expected change concerns the edition of the HTML annotations. An in-place HTML editor should be used for allowing the user to directly modify the annotation HTML, in WYSIWYG mode, without launching any external editor. Microsoft's DHTML editor is the preferred one.
In current version of the OMM, HTML annotations are viewed with Internet Explorer. That means that we can virtually handle any document that can viewed on the Web (Word document, PowerPoint presentation, VRML files, etc.).
2.2 Wordpad Annotation
A Wordpad annotation is represented by an RTF file. In the OMM, a standard “Rich Edit” component is used for visualising and editing these annotations.
2.3 Text Annotation
2.4 Closed Caption Annotation
In current version of the OMM, closed-caption annotations are created by converting a Virage VDF file into an OBVI. The user can directly create closed-caption annotation in the OMM. However, this functionality should be removed because closed-captions are supposed to be the result of an automated process (for example, during video acquisition, a hardware module grabs the closed-captions from the analog video signal). And, since closed-captions are a transcription of the speech track, allowing the user to create closed-captions will be very confusing.
2.5 SpeakerID Annotation
A SpeakerID annotation is a small line of text describing the person who is speaking. These annotations are currently created by converting a Virage VDF file into an OBVI. They can also be created manually by the user.
2.6 Database Template Annotation
Database template annotation (formerly Obvious Database Annotations, ODA) are internally represented by a description file. This text file gives the name and type definition of each field of the template.
To ensure a proper migration into the Obvious Network Technology framework, this description file will be encoded in XML. This change is expected for next major release of the OMM.
2.7 Audio Annotation
An audio annotation is constituted by a title, a description and an audio file. The audio file can be either embedded in the OVI file or located on another server. In that case, its URL is stored in the OVI file.
2.8 Object Annotation
An object annotation describes one or several object paths in the video. It has a title, a description, a URL, a video range and a list of object paths, one for each object of the annotation. Each object path is defined by a title, a description, a URL and a list of bounding boxes. The list of bounding box gives the successive positions of the object in the video sequence. It is not required to have a bounding for each frame in the range. Each bounding box is represented by the corresponding frame number, its position (x, y, width and height), a description and a URL. Text descriptions are plain ASCII.
Object annotations can be manually created or they can be the result of an automated process. In the first case, the use manually draws a bounding box around objects of interests. In the second case, the motion tracking algorithms developed by the Phoenix team will directly provide the set of bounding boxes.
Concerning the visualisation of Objet Annotations, a new module is under development. It will allow to draw the bounding boxes corresponding to a given set of object paths. This module will be implemented as an ActiveX Control, to ensure an easy integration with the OMM.
3 Publishing and Indexing an OBVI
Publishing an OBVI means converting the OBVI from the OVI file format to the database-centric file format. Thus, the publishing operation must be considered as a storage conversion: from a secondary (file-based) storage format to the primary (database-centric) storage format. Once an OBVI is published and indexed, it can be re-extracted from the database as an OVI or an XML file. Then the extracted OVI file can be modified and re-published later, creating a new OBVI version.
4 Storage Scenarios
Before discussing the details of the various storage formats, let's present some possible scenarios that show how different storage formats can be used.
In conclusion, an OBVI is stored in the primary storage format (i.e. in the database) when it is published. Then, it corresponds to an identified object that is securely managed by the database. This object has a unique identifier and is attached to a version number. Its content is indexed and client applications can do search queries for browsing and retrieving a particular OBVI. In the other hand, an OBVI is stored as an OVI file (i.e. a secondary storage format) when editing is necessary (because the user created a new OBVI or because the user wanted to save an OBVI on its local machine for further editing tasks).
The ability to export an OBVI from the database to an OVI or XML file gives a clear separation between the OBVI (as an object, independently of its storage format) and the file format that can be used for storing the OBVI outside the database.
5 Storage Format
5.1 Primary Storage Format
As previously described, the primary storage format is based upon a relational database system. The OBVI is handled as an object in the database. Concerning implementation issues, two possibilities are of interest:
The OVI file format is used by the OMM/OME/OMV suite of client applications. These applications use a low-level library (the OBVIKernel, which is an ActiveX control) for accessing and interpreting this binary file format.
An OVI can reference a video in two ways: the video file can be local (local hard-drive or on the LAN) or remote. In the second case, the OBVI is said to be bound to a registered media, meaning that a connection to the server components is necessary.
The header of the OVI file contains a number of fields. However the following fields are of special interest here because they are concern the interaction between client applications and server components.
The following tables give sample values of these field corresponding to different kind of OVI files
All the fields are empty (an null value is used to specify an empty or irrelevant field) excepting the MediaFile field which contains the full path of the media file.
In that case, VdocID and MediaID are nor relevant.
This is the case of OBVIs extracted from the database in write mode. The corresponding OVI file contains a Video Document Identifier and a Media Identifier that uniquely determine a registered media file in the OIS database. Here is a sample set of values for the header fields:
In that case, all the fields are relevant excepting the MediaLocation field.
This kind of OVI files contain a valid OBVI Identifier and a valid Version Identifier that allow to track the original OBVI and the parent version. If the user modifies this OVI file and re-publishes it, then
5.2.2 The XML File Format
The tools that allow to store an OBVI as a XML file are currently under development. However the DTD is already defined. Basically, the XML file will provide enough information to client applications for starting a runtime session. The runtime session starts by connecting to various server components. Query are sent to these runtime components for retrieving all the material needed for the visualisation/editing of the OBVI.
In current implementation, the XML file corresponding to an OBVI is used by the Obvious Java Viewer. It can be generated and distributed by the Obvious Media Server.
5.2.3 The OSF File Format
OSF stands for Obvious Streaming Format. An OSF and an OVI file contain the same information. However, the internal structure of the OSF file, composed of chunks, allows efficient streamed transmission of OBVIs. As an example of a possible use of the OSF file, a set of beta tools have been developed. They allow to build an OSF file from an OVI file, and multicast the OSF file on an IP channel. At the client level, a multicast receiver downloads the OSF streams and stores them on the client hard drive. These tools are described in more detail below in Section XVI.
OBVI version management is handled in the OIS database. Several versions of the same OBVI can be stored. Each version has a unique identifier and is represented by an author and a creation date. Different versions of the same OBVI can share the same blocks and annotations. Thus, it is possible to track changes from version to version. Version is automatically incremented during the publishing operation.
Versioning is handled by using 2 identifiers: an OBVI identifier (OBVIID) and a version identifier (VERSIONID). Different versions of the same OBVI have the same OBVIID but different values of VERSIONID.
The following gives a sample cycle of life of an OBVI. This cycle shows how versioning works and how OVI are used for extracting an OBVI from the database. It is constituted by 5 phases:
The OBVI SDK allows third-party applications to access OBVI objects from either the OVI file format, the XML file format or from the OIS database. The OBVI SDK is supposed to give the same level of functionalities than the OBVI Kernel.
In current implementation, the OBVI SDK is available as a DLL that exposes basic functions for opening, reading, editing and saving OVI files. As of writing, the OBVI SDK is internally used by the Obvious development team for the following tasks:
After this call, the handle can not be used any more.
2.1.2 Opening an OBVI from various sources
Applications using the OBVI SDK must link with the LibOBVI.lib import library. The runtime code is located in LibOBVI.dll. C/C++ applications can use the provided interface.h include file for the definition of various data types.
The following constants are used for the annotation types.
This section describes the various structures that are used by the LIBOBVI functions. Each structure has a magic number field (vMagic) and a size field (dwSize) that must be properly initialised by the application before using them in one of the LIBOBVI functions.
The magic number identifies the type of the structure. The following table gives the different possible values:
Concerning the dwSize field, it must be initialised to the size of the structure, in bytes.
This structure gives the OBVI metadata. The pProxyLocation field contains the URL of an image that acts as a graphical representation for the whole OBVI. In current implementation, the proxy image is in JPEG format.
This structure gives the Video metadata.
This structure gives binding information.
This structure describes an OBVI block.
These structures describe the OBVI block structure.
These structures describe the OBVI annotations.
The Obvious Site Directory (OSD) provides a worldwide directory service for sites. Client applications can access an Obvious Site Directory to browse available sites. The directory of sites is searchable: sites are grouped into categories and associated to a set of searchable properties.
The OSD service is hosted on Obvious technology's servers. It will reference all the companies that host an Obvious-compliant site.
2 Site Metadata and Site Categories
For each site the following metadata is stored:
Each site has a unique identifier. The value 1 is reserved for Obvious Technology's site. Values between 2 to 99 are reserved for future internal uses. Values over 100 are used to identify external sites.
Sites are grouped into categories. A site belongs to one and only one category. Categories may have subcategories. The information concerning sites and sites categories is stored in the SITE AND SITECATEGORY tables of the OIS database. Section X, entitled The Obvious Indexing System, provides more details.
Each site is associated to a list of OSM's IP addresses. OSM stands for Obvious Site Manager. It will be described in details in Section VI. The OSM is the entry point for a given site. It handles access control and replication. The list of OSM's addresses is a string containing several IP addresses separated by a sharp character (#). The string is terminated by 2 sharp characters. For example:
When a client application wants to access a specific site, it uses the first IP address in that list. In case of error (due to network congestion for example), it tries the next IP address in the list.
As of writing, two requested are implemented in the OSD: GetCategories and GetSites.
This request allows client applications to retrieve the list of site categories.
This request allows client applications to retrieve the list of sites.
Site resolution is the mechanism that client applications use for converting a Site Identifier to an IP address. For example, when an OVI file is opened with the OMM, the Site Identifier (SiteID) is extracted from the file and must be converted into a valid IP address for connecting to the appropriate backend services.
More precisely, site resolution gives a list of IP addresses, as described above. These IP addresses correspond to the different OSMs that can be used for a given site.
This site resolution mechanisms involves 2 processes
A cache lookup
A cache update
The best example of cache lookup/update is given in the Obvious Publisher. The Obvious Publisher, described in more details in section XIII entitled OVI Publishing and Indexing, is the application that allows the publishing/indexing of OVI files. It is implemented as a Wizard with 6 property pages. As shown in
The content of the tree is taken from the local cache, which contains the list of categories and site categories from the last update. The Update button allows the user to download the latest version of the sites and their categories.
Under Windows platforms, the Obvious Site Directory is implemented as an ISAPI script for Internet Information Server. The ISAPI script has a permanent connection to the OIS database, by using ADO.
A running version is available at http://odyssee.opus.obvioustech.com/OSDscript/IsapiOSD.dll
VI. Site Management
Site Management refers to the set of components and procedures that manage a given site. The goal of these components and procedures is to handle access control to resources and load balancing between redundant resources. These resources are called services. In current version, services include the Obvious Media Server and several popular video servers. These services can be replicated, meaning that a given site can track several instances of the same service, for performance and quality of service reasons.
The main module is the Obvious Site Manager (OSM). The OSM handles the underlying services and manages the access control mechanisms for the entire site. The second important module is the Obvious Load Manager (OLM). The OLM sits in the machine hosting the services and act as an interface between the OSM and these services.
In this simple case, the Obvious Media Server is replicated since it exists on 2 separate machines. The video servers are not replicated: there is only one instance of the NetShow service and only one instance of the Real Media service.
The Obvious Site Manager uses the Obvious Indexing System as a central repository for storing the list of services, their mapping on the available set of machines, security information, load balancing data, etc.
On each machine, there is one and only one Obvious Load Manager. This component may use several different DLLs called Load Handlers for interacting with the underlying service. In the example depicted in
2 Supported Services
Supported services include the Obvious Media Server and several video servers. Concerning video servers, a modular architecture allows the handling of virtually any kind of video server.
A service is defined by:
The OMS Service, described in more detail in Section VII entitled The Obvious Media Server, is the most important service that is used during runtime operations. When an OBVI is opened from a client application (the OMM/OME/OMV or the Obvious Java Viewer), requests are sent to OSM for retrieving metadata information, images, annotations, structure, etc.
The OMS Service is composed of:
The RM Service concerns the Real Server G2 video server from Real Networks.
The RM service is composed of:
The NS Service concerns the NetShow video server from Microsoft.
The NS Service is composed of:
The NST Service concerns the NetShow Theater video server from Microsoft.
The NST Service is composed of:
3 Service Replication
One of the main goal of the Obvious Site Manager is to manage service replication: a site may involve several instances of the same type of service. For example, a site may involve an OMS in Paris, London and New York. These OMSs have exactly the same features and serve the same content: they are replicated copies.
3.1 Automatic Monitoring
3.1.1 Monitoring of Service Load
The Obvious Site Manager tracks the available services and monitors their load at a regular time. Load refers here to any value that represents the current status of a service (based on the number of concurrent access, the available bandwidth, the CPU usage, etc.). The Obvious Network Architecture does not specify a generic algorithm for calculating the load of a service, since it depends on the nature of the service itself.
At a regular time, the Obvious Site Manager polls for the available services, sending them a request for retrieving their load. The retrieved load values are stored in the OIS database for logging and profiling. Four tables are used for that purpose:
These tables are described in more detail in Section X entitled The Obvious Indexing System.
3.1.2 Monitoring of Service Status
The polling schema described in previous section allows the Obvious Site Manager to get the current load of the available services. The monitoring process described in this section describes how the Obvious Site Manager gets the map of all available services. For a given site, the mapping of all available services can dynamically change. For example, if a machine hosting one or several services becomes unavailable, the Obvious Site Manager must update its service map accordingly.
The process of monitoring service status does not involve a polling mechanism. Each Obvious Load Monitor directly accesses its parent Obvious Site Manager for transmitting its new status. Status information concerns the activation/unactivation of a service. This mechanism is useful when a new service is added to the system: upon activation, the Obvious Load Manager will automatically contact its parent Obvious Site Manager for registering itself as a new available service. This facilitate installation and deployment on large site configurations.
3.2 Service Handlers
Service handlers are separate modules that handle service-specific operations. They act as an interface between the Obvious Load Manager and the services. The Obvious Load manager does not contain service-specific code for interacting with a service. Therefore, each service must correspond to a service handler.
For the OMS service, the service handler uses the OMS external interface to retrieve the current load. For the RM service, the service handler uses the Real Server SDK for interacting with the Real Server architecture. Each handler has service-specific code.
3.2.1 Service Handler for Real Server G2
Using the Real Server SDK, the service handler for Real Server G2 uses the Real Server SDK for retrieving the current load of the video server. This SDK allows to extract various properties concerning the number of connected clients, statistics about the streams, file usage etc. The current implementation of the service handler for Real Server G2 calculates the load by using the following simple formula:
NetShow Theater server is completely different from NetShow server. Its SDK gives access to several ActiveX controls that allow the management of the whole server. In particular, what the NetShow Theater SDK calls the MediaServer object, is an ActiveX control that permits client applications to retrieve useful information about the remote players, what title they are streaming, the available bandwidth, etc. The service handler for NetShow Theater internally uses this ActiveX control for calculating a global load measure.
4 Access Control
Beside replication management, the Obvious Site Manager also handles the access control to the site to which it belongs. For a given site, client application must first connect to the corresponding Obvious Site Manager for access control.
In addition to protocol-specific access control mechanisms, the Obvious Network Architecture defines its own security schema based on the following:
Site management is implemented as a set of 4 modules:
The Obvious Site Manager machine contains the ISAPI module that handles requests from client application. This ISAPI module manages access control and load balancing. The protocol used between client application and the Obvious Site Manager is described later in this chapter. The Obvious Site Manager machine also contains a NT Service that polls for the available services, retrieve their load and updates database tables in the OIS.
The three others machines contain one NT Service for the Obvious Load Manager, at least one service process (OMS, NS, RM, NST) and at least one Load Handler DLL. Each OLM NT Service handles requests from the Obvious Site Manager.
6.2 Interaction Between the OLM and the Load Handlers
The load handler of a given service must be implemented as a DLL with the following exported functions:
By calling this function, the OLM initializes the load handler. It also gets am opaque handle that contain service-specific data. TheOLM does not interpret the content of this handle and simply use it for subsequent calls to LH_GetLoad and LH_Release functions.
The OLM calls this functions before unloading the DLL from memory. It allows the load handler to free it resources.
The OLM calls this function in order to get the current load estimation of the service. The load is returned in parameter pLoad, as a long value.
6.3 Load Handler for Real Server G2
The service handler for Real Server G2 uses the Real Server for retrieving the number of connected clients. It has been implemented as a monitor plugin, as described in the Real Server SDK documentation. This monitor plugin, implemented as a DLL, registers itself for receiving monitoring information from Real Server G2.
At runtime, this DLL is loaded by 2 processes:
The communication between the two corresponding threads (that run in these two separate processes) is implemented by using a segment of shared memory. First thread, running in the RealServer process, writes load values in this shared memory. The second thread, running in the OLM process, reads these values and transmit them, on-demand, to the OSM,
The DLL file that implements the load handler for Real Server G2 is called lh_rm.dll
6.4 Load Handler for NetShow Theater
With NetShow Theater, we don't need to write a plugin for accessing the internal state of the server (bandwidth, number of clients, etc.). The NetShow Theater SDK describes a set of ActiveX objects that can used by any client application for managing and tuning the server. For the purpose of the Load Handler, the MediaServer ActiveX control is of interest because it gives direct access to the number of connected players, the available bandwidth, the description of the title being played, etc.
6.5 Load Handler for the OSM
7 Protocol between the OSM and the OLM
The Obvious Load Manager creates a TCP/IP socket on port 15600 and listens for incoming connections from the Obvious Site Manager. Each connection corresponds to one request. Each request has a code, from one of the following values:
The binary message of each request is represented by a structure. The following sections gives the details of each structure. All requests are merged in a union structure as follows:
The vRequestType must be initialised to one of the REQ_TYPE_XXX values. Depending on that value, the corresponding entry of the union structure must be properly initialised.
7.1 Getting the OLM version
The OLM version get be retrieved by using a request with the following binary format:
The vRequestType member must be initialized to REQ_TYPE_GETVERSION.
7.2 Retrieving the Current Load
For retrieving the load of a particular service, the OSM sends a request to the OLM. The binary message corresponding to this request is defined in the following structure:
The vRequestType member must be initialized to REQ_TYPE_GETLOAD.
The OLM sends back the following response:
The vLoad member represents the load value, as extracted from the service (via the corresponding service handler). The range of possible values depends on the service. However, low values must correspond to a low load. High values must correspond to a significant. Load values retrieved from different types of services can not be compared. Load values from the same type of service are compared by the Obvious Site Manager to determine the best service for a given client session.
8 Protocol Between the OLM and the OSM
An OLM can send its status to its parent OMS. This allows dynamic tracking of changes in the configuration of the services. When a new OLM is installed and deployed, it automatically register itself in the OMS, which will in turm add a corresponding entry in the OIS database. If a service stops (due to a failure or a manual operation from the administrator), the OLM sends the appropriate information to the OSM, which will update the OIS database accordingly.
8.2 Protocol Specification
8.2.1 Getting the OSM Version
The OLM can get the version of the remote OSM by sending a query whose binary structure is defined as follows:
8.2.2 Registering a Status Change
When the status of the OLM changes, it sends a message to the OLM. This message is defined by the following structure:
9 Protocol Between Client Applications and the OSM
The protocol between client applications and the OSM is built on top of the HTTP protocol. The OSM module responsible for interacting with these client applications is implemented as an ISAPI script for Internet Information Server.
Three kinds of client applications can access the OSM:
This request allows a client application to open a runtime session. After performing access control and replication management, the OSM returns the necessary information that will allow the client application to access a video server and an OMS.
This request allows client application to get some information about the Obvious Asset Manager
This request has no parameters.
This request allows client application to get some information about the Obvious Administration Server.
This request has no parameters.
VII. The Obvious Media Server
The Obvious Media Server is a server that allows client applications to view OBVIs. It is a runtime component in the sense that it is involved when an OBVI is opened and viewed from a client application. Here, “client application” refers to any application, developed by Obvious Technology (OMM/OME/OMV, Obvious Java Viewer) or a third-party, that is able to open, visualise or edit OBVIs.
The main goal of the Obvious Media Server is to serve metadata, images, structure, and annotations (it also accomplishes various others tasks that will be described later)
The Obvious Media Server acts as a service. It can be replicated and is subject to access control. Thus, it is handled by the Obvious Site Manager and has an entry in the SERVICE table of the OIS. An Obvious Load Manager can monitor its activity via a specific service handler.
An Obvious Media Server can handle several sites. This configuration is done during the installation process. An OMS internally manages a mapping between sites and DSNs, each DSN representing a connection to the OIS database of the corresponding site.
Serving metadata, structure information and annotations is a matter of extracting the information from the OIS database and sending it to the client in the appropriate format. However, serving images involves a more complex mechanism. The following sub-section will focus on that specific task.
2 Serving Images
For serving images, the Obvious Media Server uses an Image Proxy File (IPF). The IPF file contains the images that can be distributed by the OMS. When it receives a request for a specific image, the OMS determines the location of the IPF file corresponding to the desired media file (by looking at the OIS database). Then, it extracts the requested frame, scales it if necessary, and sends it to the client application.
In the simplest case, the IPF file is just the original video file. However, for improving performances, it is more efficient to build a special file, called OBF, that will be used by the OMS for extracting the requested frames. The OBF is designed to be a very efficient way for distributing video images.
In the other hand, building the OBF file is a time consuming process and implies a disk space overhead. Experiments show that the OBF file has approximately the same size than the original video file. In some cases it may be better to let the OMS use the original video file.
The current implementation of the Obvious Media Server can use both methods.
2.1 Supported Input Formats
2.1.1 Popular Formats
An IPF file is usually an AVI, a QT or an MPEG file, even if the OMS can virtually work with any video format recognised by DirectShow. In that case, the OMS works directly on the original video file and its performance depends on the video codec. For example, random access to MPEG frames is very slow. Some AVI codecs (Intel Indeo) allow very fast decompression and access time. Obvious Technology will recommend a set of codecs that should be used for the OMS. If the original video file is not encoded with one of these preferred codecs, then it is more efficient to choose the OBF file format, described below.
2.1.2 OBF Format
The OMS can also access video images from an OBF file (Obvious Backend Format). This file format has been designed for improving the performance of the Obvious Media Server. The OBF file is built from the original video file. During this build process, the user can choose the portion of the original video file that must be converted into OBF. He can also choose the image size of the output OBF file.
Allowing the user to build an OBF with a custom image size is very important. Original video files have often a high resolution and the OMS don't have to handle large images: most of the client applications that retrieve images from the OMS use them in thumbnails and 2D/3D storyboards. The image size needed in such applications is typically around 128×96.
An OBF file is a proprietary MJPEG format. It contains 3 sections: a header, an index and a body.
Concerning the header, its structure is described in the following table:
All the JPG images that constitute the OBF file have the same size, stored in the Width and Height fields. The common size of the images is chosen during the build process of the IPK file and depends on the desired image quality and the available bandwidth. A typical and recommended size is 128×96 (for a 4/3-ratio movie).
Right after the header, there is the OBF index. It is constituted by NbFrames entries. Each entry is defined by the following 12 bytes:
Concerning the size of the whole OBF file, it depends of course on the complexity of the images. A size of 5K can be achieved for a 128×96 image compressed at a 75% compression ratio. OBF files are designed to store low-resolution versions of original media files. They are designed to be efficiently used by the Obvious Media Server for distributing individual images.
2.2 Advanced Caching
The Obvious Media Server maintains a local cache of extracted images. This cache is used only when AVI, QT and MPEG files are used as the source. For OBF files, the cache is not used because the performance gain is not relevant (the extraction time of an image from an OBF file is very fast and there is no need to cache the extracted images).
An intelligent cache management is implemented: the OMS tries to predict future requests by pre-extracting images and putting them in the cache. The prediction algorithm involves to mechanisms
A cache cleanup mechanism has also been implemented. It allows to keep the cache size below a given threshold, specified during the installation process of the OMS. Each image in the cache has a counter that gives the number of times this image has been requested. After a certain amount of time, images that have a low counter value are removed from the cache.
3 Protocol Between the OMS and the OSM
There are basically two kind of interactions between the OMM and the OSM. The first one concerns security and replication. As described above in Section VI, the OMS is considered as one of the services managed by OSM's security and replication mechanisms. Each OMS has a Load Handler that is permanently polled for the load. The corresponding protocol has already been described in previous pages.
The second kind of interaction between the OMS and the OSM concerns the dispatch of session keys. The Obvious Site Manager handles access control by verifying the user credentials for a given site. Then, it calculates a session key that must be transmitted to both the client application and the Obvious Media Server. The following sections describe the protocol used between the OMS and the OSM for enabling and disabling session keys.
The protocol between the OMS and the OSM is built on top of HTTP. Encryption of transmitted data between the OMS and the OSM is handled by using a DES algorithm in CBC mode. Authentication is handled by using a N passes Zero-Knowledge algorithm.
3.2 Protocol Specification
3.2.1 Setting a New Session Key
Two kind of requests are supported by the OMS:
1) Requests concerning the runtime phase, i.e. the viewing of an OBVI:
Given a specific version of an OBVI (referenced by two identifiers: ObviID and VersionID), these requests allow client application to get all the information necessary to view the OBVI.
2) Requests concerning the retrieval of an OBVI:
Given a specific OBVI version (referenced by two identifiers: ObviID and VersionID), these requests allow client application to retrieve the OBVI in a secondary storage format (OVI or XML)1.
1 OSF extraction is not implemented in the OMS because OSF streams are supposed to be used by multicast/unicast tools such as the Obvious Multicaster and the Obvious Multicast Receiver.
4.2 Protocol Specification
The protocol between client applications and the OMS is implemented on top of HTTP.
This request allows the client to retrieve an image from a video media.
This request allows to retrieve the metadata of a given video document.
This request allows to retrieve the blocks of a given OBVI, at a specified level.
This request allows to retrieve the structure of a given OBVI.
This request allows client applications to retrieve the URL of an annotation. An annotation is represented by its identifier.
This request allows client applications to send hints to the OMS concerning the image extraction process. These hints help the OMS to update its cache and improve its performances.
This request allows to retrieve the metadata of a given OBVI.
This request allows to retrieve the metadata of a given OBVI version. The GetObviMetadata request returns
This request allows client applications to retrieve the VAMT results of a given registered Media. The user can retrieve a subset of the measures by specifying valid values for the FirstFrame and LastFrame parameters. If these parameters are both null, the whole set of measures is sent.
This request allows client applications to retrieve an OBVI as an OVI file. This operation correspond to the transformation from the primary storage to the OVI secondary storage format.
This request allows client applications to retrieve an OBVI as an XML file. This operation correspond to the transformation from the primary storage to the XML secondary storage format.
The Obvious Media Server is currently implemented for Windows NT and is constituted by 2 modules:
The NT service is responsible for extracting images from corresponding IPF files (that can either OBF, AVI, QT or MPEG files). The extraction of images is not achieved in the ISAPI script because of multithreading constraints in DirectShow. The others requests are handled by the ISAPI itself.
The ISAPI script has a cached connection to the OIS database, improving the speed of the SQL requests. ADO is used for all database operations.
A running OMS is available on http://odyssee.opus.obvioustech.com/OMSscript/oms.dll
VIII. The Obvious Administration Server
The Obvious Administration Server (OAS) allows remote administration of a given site. Administering a site is essentially a matter of modifying entries in the OIS database. Administration tools (developed by Obvious Technology or by third parties) never directly access the OIS database. They must send their requests to the OAS which is responsible for managing the database. By putting this additional layer between administration applications and the database repository, we ensure a higher level of security. We also facilitate the maintenance of the system: changes in the internal structure of the database will not have any impact on the administration tools as long as they use the standard interface of the OAS.
XML is extensively used for formatting the responses of the OAS. In particular, recordsets corresponding to data fetched from the OIS tables are formatted as XML documents and sent to the client application.
The protocol between administration applications and the OAS is built on top of HTTP. The OAS is implemented as an ISAPI script for Internet Information Server. The following pages gives the definition of all requests accepted by the OAS.
2.1 Category/Vdoc Manipulation
This request allows client applications to retrieve the list of Vdoc categories.
This request has no parameters.
This request allows client applications to add a new Vdoc category. They must specify a name, a description and the parent category identifier.
This request allows client applications to delete a Vdoc category.
The Obvious Asset Server has 2 roles:
Media files are typically created on client machines, with video acquisition cards, sound cards, closed-caption devices etc. One a media file is ready for . . .
1.2 Video Analysis
The Video Analysis and Measuring Tool (VAMT) allows fast video analysis of a media. Its current features allows automatic detection of scene changes on AVI, QT and MPEG files.
The goal of the VAMT is to analyse and gather various spatial and time related information from a video sequence. Its goal is not to find cuts. The VAMT process must be seen as a pre-processing step. The decision step is application dependant. For example, two different application may use the same results of the VAMT and interpret them differently, thus providing 2 completely different segmentations of the media.
The separation of the pre-processing step from the decision step is very important in the Obvious architecture. It ensure the reusability of the analysis processes (preserving time consuming analysis in applications where several different OBVIs may be built from the same source media). This features also allows to reinterpret at any time the measures collected during the pre-processing step, allowing for example the user to add or remove blocks. The addition and removal of blocks is simply a matter of reinterpreting the VAMT results with a different threshold.
The VAMT is designed to run on large amount of data. It is also designed to be used in parallel on multiples media sources. A special module called the VAMT Manager has been designed for handling multiple video analysis jobs. External applications (on the same machine or on remote locations) can access the VAMT Manager and perform the following tasks:
A job is defined as the process of analyzing a given media instance, from a timecode in to a timecode out. A media instance is uniquely identified by two identifiers VdocID and MediaID.
Several jobs can be ran in parallel. In addition, the current architecture defines a way for transparently using different VAMT algorithms and flavours.
The implementation involves 3 modules:
The core engine of the VAMT is implemented as an DirectShow filter. Thus, it can be used to parse any file format recognised by the DirectX Media architecture. An improved version for the Pentium III processor is available. By using SIMD instructions for the comparison of image pixels, an improvement ratio of 70% can be achieved.
The current implementation of the VAMT works on the pixel domain. It handles the decoded frame buffer of a rendering chain for its computations. Future versions of the VAMT will handle specific file formats such as MPEG for rapid extraction of spatial and/or time related information.
The DirectShow filter implementing the VAMT is called vamt.ax. The Pentium III version is available in vamtkatmai.ax. Since these filters act as COM objects, they present a custom interface that can be used from a container application to control the behaviour of the filter. This COM interface is called IVAMTCustom and is described below.
Future versions of the VAMT Engine (MPEG domain processing for example) will also be encapsulated as DirectShow filters. That will allow a complete compatibility between different VAMT Engine implementations.
Any VAMT Engine implementation must:
A container application that wants to use the VAMT Engine must first create a rendering chain containing the VAMT Engine filter.
2.2 VAMT Service
The VAMT Service is implemented as a Windows NT service, as shown in FIG. 21.
The VAMT Service has 2 operating modes.
The first one is automatic: the VAMT Service is configured to scan a given directory structure on the local file system, take all video files and analyse them. The results of the analysis are automatically stored in the OIS database. The second mode has been developed for demonstration and testing purposes. It is NOT used in normal operations. It will be described here because it gives a good understanding of the internal structure of the VAMT Service.
2.2.1 First Operating Mode (Normal)
The Windows registry contains a list of directories that must be scanned by. The administrator can edit this list of directories by using the VAMT Manager, described in next pages. Each directory is scanned for recognised media files: MPEG, AVI and MOV. Each media file must correspond to a description file. The description file is an XML file that contains various information necessary for the analysis:
The algorithm type
The description file is generated by the Obvious Management Console, when the user creates a new Media entry. The original media file and the description file are uploaded via FTP to one the Obvious Asset Manager machine.
2.2.2 Second Operating Mode (Testing Only)
The VAMT Service opens a TCP/IP socket and listens for incoming connections. The protocol used for controlling the VAMT Service is described below.
For each request sent by the client application, a TCP/IP connection must be open to the VAMT Service. On that connection, a binary-formatted message describing the request is sent. Once processed by the VAMT Service, a response message is sent back to the client application and the TCP/IP connection is dropped.
Each request is identified by a code. The possible code values are:
The following structures describe, for each type of request, the binary message that must be sent by a client application to the VAMT Service. Every structure has a vType field that contains one of the predefined REQ_constants.
Adding a New Job
The VAMT Manager is the client application that can be used for driving the VAMT Service from a remote location. It is implemented in Visual C++ with the MFC library.
As explained above, this operating mode is NOT used during normal operations. The VAMT Service is supposed to be autonomous and does not need the manual creation of analysis jobs. However, the VAMT Manager can be useful in many situations.
The list displays the current jobs. Each job is represented by:
When the user clicks on the Add button, the window in
When the user clicks on the Set Job Priority button of the main interface, the window in
X. The Obvious Indexing System
The Obvious Indexing System (OIS) is the database technology used for managing and indexing all the objects of the system: machines, video, media, video servers, OBVIs, Obvious Media Servers, Obvious Site Managers, etc. Its is global repository for registering these objects. As explained above, the OIS is the central component of the Obvious Network Architecture.
1.1 Vdoc and Media
A Video Document (Vdoc) is the format-independent concept of a video. Any physical copy of a Vdoc, in whole or in part, is called a Media, regardless of the copy's format. For example, from a Vdoc representing a TV movie, you can create 3 media:
The OIS architecture handles drop frame and non drop-frame (29.97 fps) SMPTE timecodes. For non drop-frame SMPTE, the string representation of the timecode is HH:MM:SS:FF. For drop frame SMPTE, a semi-colon is used (HH:MM:SS;FF)
1.3 Annotation and Stratification
The basic process of annotation involves the creation of a relationship between a media chunk and a description. A media chunk is described by two timecodes. The description can be a combination of:
A stratum is a logical group of annotated chunks. All annotated chunks in a stratum share a same semantic. For example, a Who stratum may be constituted by a set of annotated chunks describing the persons present in the video. The What stratum describes the objects present in the video. The process of stratification i.e. the process of creating various strata, can occur at the user interface level (manual annotation) or as a result of a computerised process (object tracking, speaker identification, etc.).
The stratification process can occur as many times as necessary, for a particular application. New strata correspond to new entries in the OIS repository. As users create annotations, new users and automated processes can select media chunks of interest with ever-increasing precision.
The OIS database is composed of 4 schemas.
2.1 Video Schema
This schema concerns the management and the cataloging of video documents and corresponding media. The OMS is the main component that uses this schema.
2.2 OBVI Schema
This schema concerns the management and the indexing of OBVIs. Published and indexed OBVIs are stored in this schema. The OMS is the main component that uses the OBVI Schema.
2.3 Access Control Schema
This schema concerns the access control facilities in the system. The OSM is the main component that uses this schema.
2.4 Replication Schema
This schema contains database objects that are related to the replication features. The OSM is the main component that uses this schema.
3 Oracle 8
The Obvious Indexing System is based on Oracle 8. Its advanced features for object management, content indexing, security and replication make it a good choice for supporting the core database technology of the OIS. Several extension modules called Cartridges can be used to add new features to the core Oracle database system. In particular, the Obvious Network Architecture extensively uses the Context Cartridge for implementing full search capabilities on OBVI annotations.
4 Database Schema Objects
The following pages describe all the database schema objects that have been defined in the OIS. These database schema objects concern the four schema previously defined . . . Users, tables and sequences.
Tables in the OIS database use only 4 built-in datatypes: NUMBER, VARCHAR2, CLOB, DATE. For more details about these type definitions, refer to the Oracle 8 documentation. Porting to another database environment should be easy for NUMBER, VARCHAR2 and DATE. Concerning the CLOB datatype it can be emulated by using a raw binary datatype.
An * symbol is used to show table columns that are part of a primary key.
4.1 Video tables
These tables are not available in all sites. They will be present on master sites, such as the Obvious Technology's site, that will host the site directory service (via the Obvious Site Directory server detailed in in more detail below in Section V).
A Sequence object, as defined by Oracle 8, is used to generate unique identifiers for various purposes. In the OIS, one Sequence object is created for each database instance. This Sequence object is called OSequence and is created with the following SQL command:
A Oracle package called OBVIPACKAGE has been created. It contains several stored procedures and functions internally used by several components.
4.7.1 Types Definitions
4.7.2 Stored Procedures
For more details on searching see Section XIV entitled OBVI Searching, two stored procedures are used: PROC_SEARCH and PROC_ADVSEARCH.
5 Installing the Obvious Indexing System
5.1 Creating the OIS database
Oracle 8 must be properly installed on the system. The following procedures describe how to create the OIS database and prepare Oracle 8 for hosting.
The easiest way to build the OIS database is to use the Oracle Database Assistant tool, provided with the standard installation of Oracle 8. The following Figures, denoted below, show how to parameter the OIS database. Another method is to use the provided SQL scripts that will automatically create and setup the OIS database.
The custom way of creating databases must be selected.
Step 2—FIG. 26:
Select the ConText Cartridge if your are installing the indexing components. The Advanced Replication option must be selected.
Step —FIG. 27:
Select the size of the data base that you require.
Step 4—FIG. 27A:
The database name is OISDB. The SID must be OIS. The internal password can be freely defined.
Step 5—FIG. 28:
Data base options window.
Step 6—FIG. 29:
File parameters window
Step 7—FIG. 30:
Step 10—FIG. 33:
5.2 Adding ConText Support
ConText support is required for database instances that handle the OBVI schema. Others schemas do not require ConText support.
The OIS distribution files are available in a self-extractable archive called oisntb1.exe Executing this file will launch the Install Shield installation program. During installation, the following parameters are required:
Name of the OIS home directory:
5.4 Creation of the OIS Schemas
There are four schemas that can be installed on a specific OIS database instance. The choice of the schemas that have to be installed depends on the nature of the particular database instance that the administrator wants to install.
For systems hosting the Obvious Site Management Component
Video & OBVI Indexing Schema
For systems hosting
It is possible to install all these schemas on the same system.
The installation of the schemas is performed by a set of four SQL scripts:
These scripts are located in $OIS_HOME.\Scripts and can be executed with the SQL Worksheet utility, provided with Oracle 8. Upon completion, these scripts create logs files in the same directory. Check them for any error.
XI. Obvious Management Console
The Obvious Management Console is the application that is used for administering the whole system. From a single graphical interface, the administrator of a site can browse for the different kinds of objects defined by the Obvious Network Architecture (Vdoc, Media, Groups, Users, Streams, Units, Services, etc.) and manage them.
These objects can be grouped into meaningful administration realms. For example, the video realm contains the Vdocs, Media and Streams objects. The security realm contains Groups and Users objects.
Each realm is graphically represented by a tree in the Obvious Management Console. Several realms can be displayed at the same time and can be dynamically added or removed.
Administration realms are not available for all user. Even if the Obvious Management Console is able to access and manage several realms, the user credentials will prohibit the access to specific realms. The installation procedure of the Obvious Management Console will also allow the configuration of the realms that can be administered from a particular machine.
2 Administration Realms
The following realms have been defined. They should cover most administration tasks in the current implementation of the system.
2.1 Site Realm
The site realm involves two kind of objects: SiteCategory and Site. A SiteCategory object may contain others SiteCategory objects and Site objects. The corresponding tree has 2 levels.
2.2 Security Realm
The security realm involves two kind of objects: Group and User. The corresponding tree has two levels. A group can contain several users and a user can be part of several groups. Depending on the preference of the administrator, the tree can show the groups at the first level and the users at the second level ore it can show the users at the first level and the groups at the second level.
2.3 Video Realm
The video realm involves four kind of objects: VdocCategory, Vdoc, Media and Stream. A VdocCategory object can contain others VdocCategory objects and Vdoc objects. A Vdoc object may contain several Media object. A Media object may contain several Stream objects. The corresponding tree can have 4 levels.
2.4 OBVI Realm
The OBVI realm involves two kind of objects: ObviCategory and Obvi objects. An ObviCategory object may contain ObviCategory objects and Obvi objects.
3 Site Map
The Obvious Management Console can also display a geographical map showing the location of the units and services involved in a specific site. As shown in
This map tool is implemented as an ActiveX control called the Obvious Map and implemented in C++. It is currently used in the Obvious Management Console but it can be embedded in any other management application.
The user interface of the Obvious Map allows the user to manually define de 2D position of each unit with a simple drag and drop operation. The corresponding geographical coordinates are stored in the Position field of the UNIT table.
When the Obvious Map is launched, it connects to the OIS database via the Obvious Administration Server described in more detail in Section VIII for retrieving the configuration of a given site, in terms of units, services and replication information.
The Obvious Management Console is currently implemented in C++/MFC. It offers an explorer-like graphical user interface: a left pane displays a hierarchy of objects and the right pane shows the details of a specific object. A new version is being implemented in VB and should offer the same level of functionalities.
Each managed object is called an AdminItem. An AdminItem has a set of properties, handles a set of child AdminItems and can display its configuration dialog. It can respond to basic events such as Configure (tells an object to display its configuration dialog), Add (tells an object to add a sub object) and Delete (tells the object to remove itself from the system). By right-clicking on an item in the left view, a contextual menu appears. For example,
Selecting an entry in the contextual menu will display a dialog box for object- specific operations. For instance, when the user selects the “Edit Media” menu entry from the contextual menu of a Media object,
The architecture of the Obvious Management Console is modular: new objects (corresponding to a new administration realm) can be easily added and administered. For that purpose, each object must be a represented by a C++ class derived from the CAdminItem class. The derived class must override some member functions.
For each administration realm, a tree is constructed. The nodes of the tree are derived from the CAdminItem class. The graphical part of the Obvious Management Console displays the trees in the left pane, dispatch the events between the objects and updates the right pane when necessary. The Obvious Management Console is completely independent of the nature of the AdminItems it displays.
XII. Video Registering
This Section will describe the steps performed at the video registering level. This concerns all the steps involved from the video acquisition to the creation of database entries for a specific Vdoc/Media. Video registering has nothing to do with OBVIs. It prepares and registers media files. Registered media files can then be used for creating and authoring OBVIs.
As explained before, these steps are accomplished from the Obvious Management Console. By right-clicking on a Vdoc category the user can create sub- categories. Then, from a Vdoc category, he creates a Vdoc. Each Vdoc is represented by a name and a description. From that Vdoc, he creates a new Media. Each Media is represented by various tags (name, description, format, standard, etc.).
The dialog box for the creation of the Media gives 2 possibilities to the user:
Video characteristics (such as the frame rate, the number of frames, the codec and the image size) are automatically extracted from the video file. Others user-defined fields (such as the name and the description of the Media) must be filled by the user.
At the end, the digital media file is processed as follows:
Then a new Vdoc/Media entry is created in the OIS database. Of course, if the Vdoc entry already exists, a new entry is appended to the list of Media entries for that Vdoc.
After the creation of the Media, the user can create Streams. By right-clicking on a Media, he launches a external tool for stream building. Typically, this tool is NetShow Encoder in the case of ASF streams or the Real producer in the case of RealMedia streams. Then, he defines the new Stream entry by its name, description, bandwidth, etc. A corresponding entry is created in the OIS database.
These constitute the only steps that must be manually accomplished from the Obvious Management Console. The others steps, described below, are performed in background, asynchronously to this first phase.
Once the original media file is uploaded to the Obvious Asset Manager, it is automatically analysed by the VAMT Service. A new analysis job is created and runs in parallel with others analysis jobs. At the end, a measures file, containing the VAMT pre-processing measures, is created and stored in the OIS database. These measures can be retrieved by any client application by sending the appropriate request to the Obvious Media Server.
XIII. OVI Publishing and Indexing
This chapter details the process of publishing and indexing OBVIs. As explained in previous chapters, an OBVI is a database object that can be exported in several forms: OVI, XML or OSF file. Currently, the only format that can support editing and authoring is the OVI file. The OMM/OME suite of tools allow the user to load an OVI file, modify it and save it locally.
The export functionality is a conversion from the promary storage format to one of the available secondary storage formats. Publishing and indexing an OVI file simply means converting an OBVI from a secondary storage format (the OVI file) to the primary (database-centric) storage format.
1 The Oracle 8 ConText Cartridge
ConText Cartridge is an Oracle extension module that gives fall text search capabilities to the Oracle 8 Server. In addition, ConText provides advanced linguistic processing of English-language text.
ConText provides advanced text searching and viewing functionality, such as full text retrieval, relevance ranking, and query term highlighting. Text queries support a wide range of search options, including: logical operators (AND, OR, NOT, etc.), proximity searches, thesaural expansion, and stored queries. Text viewing capabilities include WYSIWIG and plain text viewing of selected documents, as well as highlighting of query terms.
ConText provides in-depth linguistic analysis of English-language text. The output from this linguistic processing can be used to perform theme queries, which retrieve documents based on the main topics and concepts found in the documents.
Basically, the publishing and indexing process involves 3 major steps:
These steps are accomplished by the Obvious Publishing Engine.
2.1 Publishing the Annotations
The annotations contained in an OVI file are converted into HTML. The following table describes how this conversion is achieved, depending of the annotation type.
The conversion between Wordpad and HTML can be easily achieved by using Microsoft Word's automation features. This allows to programmatically launch a Microsoft Word application, load the Wordpad document and convert it into HTML. Microsoft Word automatically handles the conversion of the graphics and others embedded objects.
A specific converter is needed for each kind of annotation. The mapping between annotation types and converters is stored in the CONVERTER table of the OIS. For each annotation type, this table gives the GUID of the COM (or DCOM) object that can be used for processing it. A converter is a COM object that implements the IAnnotConverter COM interface, described in more detail in Section XXII entitled the IANNOTFILTER COM INTERFACE.
After being converted into HTML, each annotation is published on a Web Server and the corresponding URL is stored in the URL column of the ANNOTATION table. The publishing is achieved by doing a FTP upload on a specific directory in the remote Web Server. On this Web server, each annotation is stored in a separate directory whose name has the following syntax:
Regardless of the annotation format, the ConText Cartridge requires text to be filtered for the purposes of text indexing or text processing through the Linguistic Services (as well as highlighting the text for viewing).
Text extracted from OBVI annotations and OBVI metadata is stored in the Text column of the ANNOTATION table. Refer to section X for more details. This column stores data as a CLOB, i.e. a Character Large Object. Under Oracle 8, the CLOB data type can store single-byte text, up to 4 gigabytes in size. CLOBs have full transactional support: the CLOB value manipulations can be committed or rolled back.
At a certain time, the publishing/indexing process of an OBVI involves the extraction of text data from each annotation. The implementation details of this extraction depends on the type of the annotation. In current version of the OVI file format , the following annotations types can be found
A specific filter is used for filtering each type of annotation. To permit future extensions and enhancements, these filters are implemented as external modules (COM objects) that can be dynamically loaded and used by the Obvious Publishing Engine for retrieving text data from a given annotation. For example, when the Obvious Publishing Engine finds a Web annotation (a HTTP link to a remote HTML page), he uses a specific filter that will download the HTML code, parse it and produce raw text. All filters are supposed to output raw text that will be stored in the Content column of the ANNOTATION table. All filters present the same interface to the Obvious Publishing Engine, FIG. 39.
The mapping between annotation types and filters is stored in the FILTER table of the OIS. For each annotation type, this table gives the GUID of the COM (or DCOM) object that can be used for processing it. A filter is a COM object that implements the IAnnotFilter COM interface, described in more detail below in Section XXI entitled GUID For Objects.
2.3 Creating Database Entries
The database format is the primary storage format for an OBVI. An OBVI is uniquely represented by 2 identifiers: the OBVI Identifier and the Version Identifier. If the OVI file is already bound to an OBVI in the database then the publishing process consists of creating a new version. Otherwise, a new OBVI (with a starting version identifier) is created. The creation of database entries for a new OBVI involves the manipulation of several database tables. First, a new OBVI Identifier and a new Version Identifier are allocated (see OBVI and VERSION tables). Then, new Blocks are created (see BLOCK table) and bound to the OBVI (see BLOCKFOROBVI table). Finally, new Chunks are created (see CHUNK table) and bound to the OBVI (see CHUNKFOROBVI table). The ANNOTATION table.
In the case of publishing a new version of the OBVI (rather than publishing a new OBVI), the procedure is roughly the same. The only difference concerns the reuse of Blocks, Chunks and Annotations. As explained before, since changes between versions are supposed to be small (the user typically changes adds or removes some blocks and edit few annotations) the system tries to reuse Blocks, Chunks and Annotations from the previous version.
Three modules are implemented:
The most important module is the Obvious Publishing Engine, responsible for the publishing and indexing process. It internally uses a set of filters for gathering text information from the various kinds of annotations found in the OVI file. It also uses a set of converters for transforming OVI annotations into HTML.
The following filters have implemented. As explained before, they are COM objects that implement the IAnnotFilter interface.
Under Windows NT, the Obvious Publishing Engine is implemented as a NT service. It scans a list of predefined directories. For every OVI file found, the Obvious Publishing Engine starts an indexing process (a new thread). Several OVI files can be indexed at the same time.
The code of the core indexing process is located in a DLL called LibINDEX.dll. This DLL contains several exported functions but the most important one is called LIBINDEX_IndexOVI. This function accomplishes all the necessary steps for publishing and indexing an OVI file.
Most of the code uses ADO for accessing and updating the various tables of the OIS database. It also uses the OCI library for Oracle specific code concerning the handling of CLOB data.
3.4 Obvious Publishing Manager
As described in previous pages, the Obvious Publishing Engine can be controlled by a client application, by using a TCP/IP connection. The Obvious Indexing Manager is a sample of such client application. It is implemented in C++/MFC. It opens a TCP/IP connection to the machine hosting the Obvious Publishing Engine and sends requests for:
The Obvious Publishing Manager is a configuration tools that can be used by administrators for controlling and tuning the Obvious Publishing Engine.
3.5 Obvious Publisher
The Obvious Publisher is the graphical interface from which a user launches the publishing and indexing of its OVI files. The Obvious Publisher is supposed to run on a client machine, where OVI files are located.
The Obvious Publisher is implemented as a Wizard encapsulated in an ActiveX Control. It has been developed in C++/MFC. This ActiveX Control has only one automation function: RunWizard. A container application can call this function to launch the Wizard. It has the following steps:
After these steps, the OVI file is now on the machine where the Obvious Publishing Engine is located. The Obvious Publishing Engine will automatically handle all the steps for publishing and indexing the OVI. It will parse the annotations, extract raw text for indexing purposes, convert them into HTML and publishe these annotations on Web servers. It will also create database entries for the new OBVI version.
By using the Obvious Publisher wizard, the user can send several OVI files for publishing. They will be handled by the Obvious Publishing Engine in batch. The Email address that the user entered in the fourth page of the Wizard is used by the Obvious Indexing Engine for sending any error report to the author.
XIV. OBVI Searching
Under current implementation, search capabilities are provided by the Context Cartridge engine. As explained before, on of the task accomplished by the Obvious Publishing engine is the filtering of annotations: the OVI annotations are extracted and filtered to produce raw text that can be indexed by the ConText Cartridge. This raw text is stored in the Text column of the ANNOTATION table.
The ConText Cartridge has its own indexing servers. They run in background and they continuously update the internal index if the content of the Text column changes. This chapter will focus on using the search capabilities of the ConText Cartridge to build a global search platform in the Obvious Network Architecture.
Two search methods have been implemented: the basic search and the advanced search.
1.1 Basic Search
The basic search procedure allows the user to enter a keyword (or a list of keyword). This keyword is searched in every annotation, for all OBVIs.
1.2 Advanced Search
The advanced search procedure allows the user to enter different keywords for different strata.
2.1 Stored Procedures for Searching
For performance reasons, the search code has been implemented as a set of Oracle stored procedures, written with the PL/SQL language. These stored procedures are part of the OBVIPACKAGE package2. Two procedures are of interest: PROC_SEARCH and PROC_ADVSEARCH. They corresponding to the basic search and the advanced search mechanisms respectively.
2The OBVIPACKAGE package contains all the Oracle stored prodecures that have been implemented in the OIS.
The PROC_SEARCH Procedure
Given a Category Identifier and a keyword (or a list of keywords), this function runs the Context Cartridge's search engine for finding all the annotations (in all indexed OBVIs) that are in the specified category and that contain the specified keyword.
FUNC_ISCHILDOF is another function of the OBVIPACKAGE package. This helper function determines the parent/child relationship between two categories.
TCursor is an Oracle cursor type. Its definition is given in the OBVIPACKAGE package definition.
The PROC_ADVSEARCH procedure:
Given a keyword (or a list of keywords) for each strata, this functions runs the Context Cartridge's search engine on each strata, for a given category.
2.2 Search Pages
The search pages (for basic and advanced search) have been written as ASP pages. Although some of these pages use ADO for accessing the OIS database, they do not use ADO for executing a search request. Searches are handled by an Active Server Object called Obvious Search Engine. This module has been implemented in C++/ATL and contains the Oracle-specific code necessary for calling the PROC_SEARCH stored procedure responsible for the search3.
3Calling Oracle stored prodecures from ADO is tricky: The ObviousSearchEngine uses the OCI library for direct access to all Oracle features.
The latest version of the ASP search pages can be seen at http://odyssee.opus.obvioustech.com/XXX
The first page allows the user to choose between the basic search and the advanced search, FIG. 47.
From the given list of keywords, a request is sent to the database, via the Obvious Search Engine. Results are grouped by OBVI and by version. For each OBVI version, a list of chunks (timecodes) shows the exact location of the hits.
By clicking on the image, the corresponding video is played. In
By clicking on the Version ID field, the OBVI is downloaded in an OVI form. For that purpose a GetObviAsOvi request is sent the OMS. The OMS sends backs the OVI file corresponding to the specific OBVI version. In next version, the user will also be able to click on a chunk. In that case, the OVI will be downloaded and the OMM will automatically position itself on that specific chunk.
XV. OBVI Indexing with MIS
1 Microsoft Index Server
Microsoft® Index Server is a full-text indexing and search engine for Microsoft Internet Information Server (IIS) and Microsoft Windows NT.RTM. Server. It allows any Web browser to search documents for key words, phrases, or properties such as an author's name.
Index Server is designed for use on a single Web server on an intranet or the Internet. It can easily handle large numbers of queries on a busy site. Automatic updating and support for Microsoft Office documents is ideal for an intranet where files change frequently.
Index Server is capable of indexing textual information in any document type through content filters. Filters are provided for HTML, text, and Microsoft Office documents. Application developers can provide support for any other document by writing to the open IFilter interface. An IFilter knows how to read a file and extract the text. This text can then be indexed.
2 Indexing OBVIs with MIS
MIS uses catalogs for storing the index information related to a set of directories. By default, the Web catalog is bound to the root hierarchy of the local Web site. The administrator of the system can create others catalogs. It is recommended to create
An MIS filter has been implemented in C++. It allows MIS's indexing engine to parse OVI files and gather useful information for indexing. This filter basically implements the IFilter COM interface and internally uses the OBVI SDK, described in more detail in Section IV, for opening and reading OVI files.
The filter must registered on the system. Then, any OVI file present will be automatically indexed by MIS. Once indexed, queries can be ran from a web browser or any MIS-compliant application.
can be fetched ch resultscan either create a new MIS catalog or use the pre- defined Web catalog.
As described above, OSF is, with OVI and XML, another secondary storage format. OBVIs saved as OSF files can be efficiently streamed. This Section will focus on the specific tools that have been developed for building OSF files, streaming OSF data over IP multicast channels and receiving channels content at the client side.
1 The OSF Specification
An OSF file is composed by several chunks: the metadata chunks, the structure chunks, the image chunks and the annotation chunks. Each chunk is encoded with several data packets. A packet has the following binary structure:
The pSync field allows the client applications to parse asynchronous OSF streams and synchronise the OSF reading.
Several OSF can be transmitted on the same communication channel. In that case, packets corresponding to different OSFs can be interleaved. The vOsfID allows the identification of each packet. It allows client applications to group received packets by OSF and rebuild the original stream.
The chunk type, stored in the vType field, can be one of the following:
The vDataSize field gives the number of bytes that constitute the packet data. This data starts at the pData field.
Each chunk type can be transmitted by using several packets. In that case, the vNumPacket and vNbPacket allow client application to reconstruct the original data chunk. This is useful with UDP protocols for example, where the maximum block of data that can be transmitted at each call is limited.
2 The Obvious Stream Builder
The Obvious Stream Builder is simple tool that allows the conversion of OVI files into OSF files. It internally uses the OBVI SDK (LibOBVI.dll) for parsing the input OVI file and creating corresponding packets for the OSF file.
3 The Obvious Multicaster
The New button allows the user to create a new channel.
4 The Obvious Multicast Listener
XVII—The whole picture
This section gives an overview of the whole process. This process concerns the following tasks:
From the same Obvious Administration Console, the user starts the Obvious VAMT Manager. Then he creates and launches VAMT analysis jobs for the media that he has just created.
Cycle 2: Creating and Authoring OBVIs from Pre-registered Media
In the Obvious Server Architecture, many HTTP requests give a response that can be represented as a recordset, i.e. a table constituted by N fields and M rows. Each field has a type and a name.
An XML format has been designed for representing a generic recordset. This allows a common representation of all these HTTP responses.
The DTD is given below:
The GetStructure request of the Obvious Media Server returns a XML-formatted response whose DTD is described below:
A sample XML is given below. It represents a structure with 3 levels. First level is composed of 2 blocks, with 2 child blocks each.
Object annotations are internally represented by an XML file whose DTD is described below:
A sample XML file is given below.
A Global Unique Identifier (GUID), also called Universal Unique Identifier (UUID), is a 128-bit value used in cross-process communication to identify entities such as client and server interfaces, manager entry-point vectors, and RPC objects.
As previously described, the Obvious Network Architecture defines unique identifiers for various objects, such as Vdocs, Media, Streams, Users, Groups, Units, OBVIs, Versions, etc. However, these identifiers are not unique over all sites. Two objects from two different sites may have the same identifier.
This section describes a way for creating global unique identifier. These GUIDs would permit the referencing of objects across sites boundaries making possible for one site to access the objects of another site.
The structure of a GUID is given in the following matrix:
Suppose we have 2 Sites called A and B. From a client application a user fetches an object XA from site A. This object has a unique identifier in Site A. However it is not guaranteed that this identifier is not already in use in Site B. A GetGUID request is sent to the OSM of Site A. This allows the client application to get the GUID corresponding to XA.
XXII. The IAnnotFilter COM Interface
The Obvious Publishing Engine uses a set of filters for extracting raw text from the various kind of annotations that can be found in an OVI. Each OVI annotation correspond to a specific filter that acts as a parser for that annotation. The Obvious Network Architecture specifies that filters are COM objects that implements the IAnnotFilter interface. This COM interface is described below.
XXIII. The IAnnotConverter COM Interface
During the publishing/indexing process, the Obvious Publishing Engine uses a set of converters for converting OVI annotations into HTML. Each OVI annotation must be handled by a specific converter. The Obvious Network Architecture specifies that converters are COM objects that implements the IAnnotConverter interface. This COM interface is described below.