Processing of video content Number:7,167,809 from the United States Patent and Trademark Office (PTO) owispatent Processing of video content including sports.<H Processing, content, Processing, of, v, 7167809.html, Patent, pto, 7167809

Title: Processing of video content

Abstract: Processing of video content including sports.

Patent Number: 7,167,809 Issued on 01/23/2007 to Li

Inventors: Li; Baoxin (Vancouver, WA)
Assignee: Sharp Laboratories of America, Inc. (Camas, WA)

Appl. No.: 10/976,578

Filed: October 30, 2004

Related U.S. Patent Documents


Application Number	Filing Date	Patent Number	Issue Date
10404987	Mar., 2003	7006945

Current U.S. Class: 702/181

Current International Class: G06F 15/00 (20060101)

Field of Search: 702/181,182-185 345/461,481,473

References Cited [Referenced By]

U.S. Patent Documents


4183056	January 1980	Evans et al.
5521841	May 1996	Arman et al.
5574845	November 1996	Benson et al.
5805733	September 1998	Wang et al.
5828809	October 1998	Chang et al.
5923365	July 1999	Tamir et al.
5959681	September 1999	Cho
5995095	November 1999	Ratkonda
6439572	August 2002	Bowen
2004/0197088	October 2004	Ferman et al.

Other References

Boreczky, John S. and Wilcox, Lynn D., A Hidden Markov Model Framework for Video Segmentation Using Audio and Image Features., FX Palo Alto Laborator article, four pages. cited by other .
Kobla, Vikrant, et al., Detection of Slow-Motion Replay Sequences for Identifying Sports Videos, Laboratory for Language and Media Processing, University of Maryland article, six pages.2. cited by other .
Intille, Stephen S., Tracking Using a Local Closed-World Assumption: Tracking in the Football Domain, partial masters thesis submitted to Media Arts & Sciences, School of Architecture and Planning Aug. 5, 1994, consisting of 62 pages. cited by other .
Kobla, Vikrant, et al., Identifying Sports Video Using Reply, Text, and Camera Motion Features, Laboratory for Language and Media Processing, University of Maryland, consisting of 12 pages. cited by other .
Levinson, S.E., Rabiner, L.R. and Sondhi, M.M., An Introduction to the Application of the Theory of Probablistic Functions of a Markov Process to Automatic Speech Recognition, The Bell System Technical Journal, Vo. 62, No. 4, Apr. 1983, pp. 1035-1074. cited by other .
Saur, Drew D., et al., Automated Analysis and Annotation of Basketball Video, SPIE vol. 3022, pp. 176-187. cited by other .
Yow, Dennis, et al., Analysis and Presentation of Soccer Highlights from Digital Video, Asian Conference on Computer Vision, 1995, consisting of five pages. cited by other .
Golin, Stuart J., New Metric to Detect Wipes and Other Gradual Transitions in Video, SPIE vol. 3656, Jan. 1999, pp. 1464-1474. cited by other .
Courtney, Jonathan D., Automatic Video Indexing Via Object Motion Analysis, Pattern Recognition, vol. 30, No. 4, pp. 607-625, 1997. cited by other.

Primary Examiner: Raymond; Edward
Attorney, Agent or Firm: Chernoff, Vilhauer, McClung & Stenzel

Parent Case Text

CROSS-REFERENCE TO RELATED APPLICATIONS

This application is a division of U.S. patent application Ser. No. 10/404,987, filed Mar. 31, 2003, now U.S. Pat. No. 7,006,945, which claims the benefit of Provisional App. No. 60/439,240, filed Jan. 10, 2003.

Claims

The invention claimed is:

1. A method of automatically indexing a video of a football game having a plurality of plays, said video comprising a plurality of segments, each segment having at least one frame, where said plurality of segments are arranged in a sequence of types of shots comprising a shot of a scoreboard that precedes a play, a shot of said play from a sideline, and a shot of said play from an end zone, said method comprising: (a) inputting a said shot to a computing device; and (b) said computing device categorizing said shot as one of a scoreboard shot preceding said play, a sideline shot of said play, or an end zone shot of said play by: (i) categorizing said shot as a scoreboard shot based on at least one of: (1) whether said shot has less than a threshold of green pixels; (2) said shot is of less than a threshold duration; and (3) whether said shot has a motion complexity less than a threshold; and (ii) categorizing said shot as either an end zone shot or a sideline shot based on at least one of: (1) whether the preceding shot was classified as a scoreboard shot; and (2) whether an estimated angle of sidelines in a shot exceeds a threshold.

2. The method of claim 1 including the step of said computing device indexing said shot by a play identifier and a type of shot.

3. The method of claim 1 where said plurality of shots are arranged in a sequence of a shot of a scoreboard that precedes a play, followed by a shot of said play from a side line, followed by a shot of said play from an end zone.

4. The method of claim 1 where said plurality of shots are arranged in a sequence of a shot of a scoreboard that precedes a play, followed by a shot of said play from an end zone, followed by a shot of said play from a sideline.

5. The method of claim 1 where said sequence includes a shot of said play from a first sideline and a shot of said play from a second side line.

6. A method of automatically indexing a video of a football game having a plurality of plays, said video comprising a plurality of segments, each segment having at least one frame, where said plurality of segments are arranged in a sequence of types of shots comprising a shot of a scoreboard that precedes a play, a shot of said play from a sideline, and a shot of said play from an end zone, said method comprising: (a) a computing device detecting a shot; and (b) said computing device classifying said shot as one of a scoreboard shot preceding a said play, a sideline shot of a said play, or an end zone shot of a said play using a Hidden Markov Model.

7. The method of claim 6 where said step of categorizing said shot comprises using a Viterbi algorithm to find the most likely shot-type sequence.

8. The method of claim 6 where said Hidden Markov Model is a trained Hidden Markov model established by (i) labeling each frame in a plurality of shots in a training sequence by shot-type; (ii) computing a feature vector for each frame in said training sequence; and (iii) using a Baum Welch algorithm to estimate model parameters for said Hidden Markov Model.

9. The method of claim 8 where said classification is done by a first-order Markov transition model.

10. The method of claim 9 where said Hidden Markov model uses a Bayesian rule to estimate the probability that a current shot is of a respective shot type given the type of previous shot and the feature vector of the current frame, based on the probability of a feature vector given the type of shot for the current frame, and a first order transition matrix associating the probability of a shot type of a current frame based on the shot type of a the previous frame.

11. The method of claim 10 where said feature vector is defined by a first component comprising the percentage of green pixels in the frame, a second component comprising a color histogram difference between the first and last frame of the shot, and a third component comprising the length of the shot, and where said first-order transition model computes the probability of said feature vector given the shot type for the current frame based on the independent probabilities of the respective components given the shot type o fteh current frame.

Description

BACKGROUND OF THE INVENTION

The present invention relates to processing of video content.

The amount of video content is expanding at an ever increasing rate, some of which includes sporting events. Simultaneously, the available time for viewers to consume or otherwise view all of the desirable video content is decreasing. With the increased amount of video content coupled with the decreasing time available to view the video content, it becomes increasingly problematic for viewers to view all of the potentially desirable content in its entirety. Accordingly, viewers are increasingly selective regarding the video content that they select to view. To accommodate viewer demands, techniques have been developed to provide a summarization of the video representative in some manner of the entire video. Video summarization likewise facilitates additional features including browsing, filtering, indexing, retrieval, etc. The typical purpose for creating a video summarization is to obtain a compact representation of the original video for subsequent viewing.

There are three major approaches to video summarization. The first approach for video summarization is key frame detection. Key frame detection includes mechanisms that process low level characteristics of the video, such as its color distribution, to determine those particular isolated frames that are most representative of particular portions of the video. For example, a key frame summarization of a video may contain only a few isolated key frames which potentially highlight the most important events in the video. Thus some limited information about the video can be inferred from the selection of key frames. Key frame techniques are especially suitable for indexing video content but are not especially suitable for summarizing sporting content.

The second approach for video summarization is directed at detecting events that are important for the particular video content. Such techniques normally include a definition and model of anticipated events of particular importance for a particular type of content. The video summarization may consist of many video segments, each of which is a continuous portion in the original video, allowing some detailed information from the video to be viewed by the user in a time effective manner. Such techniques are especially suitable for the efficient consumption of the content of a video by browsing only its summary. Such approaches facilitate what is sometimes referred to as "semantic summaries".

The third approach for video summarization is manual segmentation of the video. In this manner each portion of the video that is determined to be of interest is selected. The selected segments of the video are then grouped together to form a video sequence comprising the selected segments. In some cases a new video sequence is constructed from the selected segments, and in other cases the segments are identified in the existing sequence so that they may be viewed in sequence (while not viewing the non-selected segments).

What is desired, therefore, is a video processing technique suitable for video.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1A is an exemplary flowchart for play and non-play detection.

FIG. 1B is an exemplary flowchart for play detection.

FIG. 2 is an exemplary illustration of a hiking scene in football.

FIG. 3 is an exemplary illustration of a kicking scene in football.

FIG. 4 illustrates one example of a generally green color region.

FIG. 5 is a technique for defining the generally green color region.

FIG. 6 is a technique for defining histograms for the field frames.

FIG. 7 illustrates the definition of a central region of a frame and/or field.

FIG. 8 illustrates candidate frame selection based upon an initial generally green selection.

FIG. 9 is an exemplary illustration of a hiking scene in football.

FIG. 10 illustrates edge detection for the image in FIG. 9.

FIG. 11 illustrates parametric lines for the edge detection of FIG. 10.

FIG. 12 illustrates computed motion vectors for football video.

FIG. 13 illustrates an exemplary start of a football play.

FIG. 14 illustrates a green mask for the image of FIG. 13.

FIG. 15 illustrates an exemplary green mask for an image of a football player.

FIG. 16 illustrates an exemplary football player.

FIG. 17 illustrates a projection of the green mask of FIG. 14.

FIG. 18 illustrates a projection of the green mask of FIG. 15.

FIG. 19 is an illustration of temporal evidence accumulation.

FIG. 20 is an illustration of the U-V plane.

FIG. 21A illustrates changes between frames.

FIG. 21B is an illustration of detecting the end of a play in football.

FIGS. 22A 22F illustrates the start of a baseball play.

FIG. 23 illustrates one technique for play detection for baseball.

FIG. 24 illustrates a temporal frame validation technique.

FIG. 25 illustrates color selection for plays.

FIG. 26 illustrates the frame breaks between plays.

FIG. 27 is an exemplary flow chart for determining the end of a play.

FIG. 28 illustrates three constituent shots of the same play (from left to right): scoreboard, sideline shot, and end-zone shot.

FIG. 29 illustrates components of a deterministic approach FIG. 30 illustrates abnormal frame (logo overlay on the field, shadow, distorted color, etc) for using percentage of green pixels in detecting an SL/EZ shot.

FIG. 31 illustrates empirical length distributions of SB and SL/EZ, respectively.

FIG. 32 illustrates empirical D distributions for SB and SL/EZ, respectively, showing significant overlaps.

FIG. 33 illustrates a 4 state Hidden Markov Model.

FIG. 34 illustrates a simple first-order Markov transition model for modeling the transitions between SB, SL, and EZ.

FIG. 35 illustrates a system embodying aspects of the present invention.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT

Sumo Wrestling

Sumo, the national sport of Japan, is tremendously popular in eastern Asia and is growing in popularity elsewhere in the world. Sumo is a sport comprising bouts in which two contestants meet in a circular ring 4.55 meters in diameter. The rules of Sumo are uncomplicated. After the contestants and a referee have entered the circular ring, the bout begins with an initial charge--called a "tachiai"--where each contestant rushes towards, then collides with, the other. The bout will end when one of the contestant loses by either stepping outside the circular ring or touching the ground with any part of the contestant's body other than the soles of the feet. Aside from a limited number of illegal moves, such as gouging the opponent's eyes, striking with a closed fist, or intentionally pulling at the opponent's hair, there are no rules that govern a sumo bout.

Sumo participants may compete against each another in one of a number of tournaments. Japan sponsors six sanctioned Grand Sumo tournaments, held in odd-numbered months throughout the year, in which competitive sumo contestants face one another with the opportunity for advancement in rank. Sumo contestants are ranked under a strict meritocracy; winning bouts in these sanctioned tournaments improves a competitor's rank while losing bouts diminishes that rank. Aside from the six sanctioned tournaments, a number of exhibition tournaments--called Jungyo--are scheduled throughout the year.

Though a sumo tournament will typically take place over several weeks with bouts scheduled throughout each day, most bouts of interest, i.e. those involving higher ranked contestants, are scheduled to begin late afternoon when live television broadcasts of the tournament occur. These portions of the sumo tournaments usually last 2 3 hours each day and are often video recorded for later distribution or for re-broadcast.

Though such a video of a sumo tournament might typically last about 2 3 hours, only about ten minutes turns out to include time during which two players are in a bout. An individual sumo bout is brief; the typical bout will end with the initial collision, though a rare bout might last two to three minutes. Interspersed between bouts are a large number of ceremonies that precede and follow each bout.

Though brief, the time intervals during which a bout is proceeding are intense and can captivate those in the viewing audience, many of whom are able to identify a myriad of named sumo techniques that may occur in rapid succession. Such techniques include a "kekaeshi" (a foot-sweep), a "kubinage" (a head-lock throw), and an "izori" (a technique where a contestant crouches below the opponent's rush, grabbing one of the opponent's legs, lifting the opponent upon the shoulders and falling backwards), as well as some sixty five to seventy more named sumo techniques or occurrences.

The remaining time during the sumo tournament is typically not exciting to watch on video. Such time would include for example inter-bout changes of players, pre-bout exercises and ceremonies, post-bout ceremonies and in the case of broadcast, nearly endless commercials. While it may indeed be entertaining to sit in an arena for several hours for a sumo tournament, many people who watch a video of a sumo tournament find it difficult to watch all of the tournament, even if they are rabid fans. Further, the tournaments are held during daytime hours, hence many fans are unable to attend a tournament or to watch a live broadcast due to work. Such fans may nonetheless be interested in watching specific bouts or some other condensed version of the tournament. Thus a video summarization of the sumo tournament that provides a summary of the tournament having a duration shorter than the original sumo video, may be appealing to many people. The video summarization should provide nearly the same level of the excitement (e.g. interest) that the original game provided.

Upon initial consideration, sumo would not be a suitable candidate to attempt automated video summarization. Initially, there are nearly an endless number of potential moves that may occur that would need to be accounted for in some manner. In addition, each of these moves may involve significant player motion that is difficult to anticipate, difficult to track, and is not consistent between plays. In addition, the players are flesh toned and the ring is likewise generally flesh toned making identification of the events difficult. Based upon such considerations it has been previously considered impractical, if not impossible, to attempt to summarize sumo.

It is conceivably possible to develop highly sophisticated models of a typical sumo video to identify potentially relevant portions of the video. However, such highly sophisticated models are difficult to create and are not normally robust. Further, the likelihood that a majority of the highly relevant portions of the sumo video will be included in such a video summarization is low because of the selectivity of the model. Thus the resulting video summarization of the sumo tournament may simply be unsatisfactory to the average viewer.

Baseball

A typical baseball game lasts about 3 hours of which only about one hour turns out to include time during which the ball is in action. The time during which the ball is in action is normally the exciting part of the game, such as for example, pitching the ball to the batter, hitting a home run, hitting the ball, running the bases, a pitch to first base, pitching a "foul" ball, pitching a "strike" ball, pitching a "ball", fouling the ball to the bleachers, catching a pop fly, etc. The remaining time during the baseball game is typically not exciting to watch on video, such as for example, nearly endless commercials, the time during which the players change from batting to fielding, the time during which the players walk to the plate, the time during which the players walk around between innings, the time during which the manager talks to the pitcher, the time during which the umpire cleans home plate, the time during which the batter swings the bat in practice, the time during which the batter just waits for the pitcher, the time during which the spectators are viewed in the bleachers, the time during which the commentators talk, etc. While it may indeed be entertaining to sit in a stadium for three hours for a one hour baseball game, many people who watch a video of a baseball game find it difficult to watch all of the game, even if they are loyal fans. A video summarization of the baseball video, which provides a summary of the game having a duration shorter than the original baseball video, may be appealing to many people. The video summarization should provide nearly the same level of the excitement (e.g. interest) that the original game provided.

It is possible to develop highly sophisticated models of a typical baseball video to identify potentially relevant portions of the video. However, such highly sophisticated models are difficult to create and are not normally robust. Further, the likelihood that a majority of the highly relevant portions of the baseball video will be included in such a video summarization is low because of the selectivity of the model. Thus the resulting video summarization of the baseball game may simply be unsatisfactory to the average viewer.

Football

A typical football game lasts about 3 hours of which only about one hour turns out to include time during which the ball is in action. The time during which the ball is in action is normally the exciting part of the game, such as for example, a kickoff, a hike, a pass play, a running play, a punt return, a punt, a field goal, etc. The remaining time during the football game is typically not exciting to watch on video, such as for example, nearly endless commercials, the time during which the players change from offense to defense, the time during which the players walk onto the field, the time during which the players are in the huddle, the time during which the coach talks to the quarterback, the time during which the yardsticks are moved, the time during which the ball is moved to the spot, the time during which the spectators are viewed in the bleachers, the time during which the commentators talk, etc. While it may indeed be entertaining to sit in a stadium for three hours for a one hour football game, many people who watch a video of a football game find it difficult to watch all of the game, even if they are loyal fans. A video summarization of the football video, which provides a summary of the game having a duration shorter than the original football video, may be appealing to many people. The video summarization should provide nearly the same level of the excitement (e.g. interest) that the original game provided.

Upon initial consideration, football would not be a suitable candidate to attempt automated video summarization. Initially, there are nearly an endless number of potential plays that may occur which would need to be accounted for in some manner. Also, there are many different types of plays, such as a kickoff, a punt, a pass play, a kickoff return, a running play, a reverse play, an interception, a sack, etc., that likewise would need to be accounted for in some manner. In addition, each of these plays involves significant player motion which is difficult to anticipate, difficult to track, and is not consistent between plays. Moreover, the ball would normally be difficult, if not impossible, to track during a play because much of the time it is obscured from view. For example, it would be difficult to distinguish interesting play related activity from typical pre-play activity of the players walking around the field getting ready for the next play. Based upon such considerations has been previously considered impractical, if not impossible, to attempt to summarize football.

It is conceivably possible to develop highly sophisticated models of a typical football video to identify potentially relevant portions of the video. However, such highly sophisticated models are difficult to create and are not normally robust. Further, the likelihood that a majority of the highly relevant portions of the football video will be included in such a video summarization is low because of the selectivity of the model. Thus the resulting video summarization of the football game may simply be unsatisfactory to the average viewer.

Video Content Including Sports

It may be observed that many different types of video content, including for example sporting events, include a game or activity that lasts a significant period of time of which only a relatively short duration of which turns out to include time during which interesting activity is occurring. The time during which interesting action is occurring is normally the exciting part of the game, such as for example, a kickoff, a hike, a pass play, a running play, a punt return, a punt, a field goal, etc. The remaining time during the video content is typically not exciting to watch on video, such as for example, nearly endless commercials, the time during which the players change from offense to defense, the time during which the players walk onto the field, the time during which the players are in the huddle, the time during which the coach talks to the quarterback, the time during which the yardsticks are moved, the time during which the ball is moved to the spot, the time during which the spectators are viewed in the bleachers, the time during which the commentators talk, inter-bout changes of players, pre-bout exercises and ceremonies, post-bout ceremonies, the time during which the players change from batting to fielding, the time during which the players walk to the plate, the time during which the players walk around between innings, the time during which the manager talks to the pitcher, the time during which the umpire cleans home plate, the time during which the batter swings the bat in practice, the time during which the batter just waits for the pitcher, the time during which the spectators are viewed in the bleachers, the time during which the commentators talk, etc. While it may indeed be entertaining to watch for several hours for a one hour activity, many people who watch a video of a sporting event find it difficult to watch all of the event, even if they are loyal fans. A video summarization of the video, such as sporting videos, which provides a summary of the event having a duration shorter than the original video, may be appealing to many people. The video summarization should provide nearly the same level of the excitement (e.g. interest) that the original game provided.

As previously discussed, upon initial consideration, sporting events would not be a suitable candidate to attempt automated video summarization. Initially, there are nearly an endless number of potential plays that may occur which would need to be accounted for in some manner. Also, there are many different types of plays, that likewise would need to be accounted for in some manner. In addition, each of these plays involves significant player motion which is difficult to anticipate, difficult to track, and is not consistent between plays. Moreover, any balls or other items would normally be difficult, if not impossible, to track during a play because much of the time it is obscured from view. For example, it would be difficult to distinguish interesting play related activity from typical pre-play activity of the participants walking around getting ready for the next play. Based upon such considerations has been previously considered impractical, if not impossible, to attempt to summarize sporting events.

It is conceivably possible to develop highly sophisticated models of a typical activity to identify potentially relevant portions of the video. However, such highly sophisticated models are difficult to create and are not normally robust. Further, the likelihood that a majority of the highly relevant portions of the video will be included in such a video summarization is low because of the selectivity of the model. Thus the resulting video summarization of the event may simply be unsatisfactory to the average viewer.

Play Selection

After consideration of the difficulty of developing highly sophisticated models of a video to analyze the content of the video, as the sole basis upon which to create a summarization, the present inventors determined that this technique is ultimately flawed as the models will likely never be sufficiently robust to detect all the desirable content. Moreover, the number of different types of model sequences of potentially desirable content is difficult to quantify. In contrast to attempting to detect particular model sequences, the present inventors determined that the desirable segments in terms of understanding, following, or even appreciating the game is limited. These important portions occur semi-periodically and sparsely during the game, but they contain the moments of intense action and are the essence of a game. The remaining time is typically less important. Therefore preferably the events are selected based upon a "play". A "play" may be defined as a sequence of events defined by the rules of the event. In particular, and in one aspect, the sequence of events of a "play" may generally include the time between which the players set up to start an activity and the time during which the activity is completed. A play may also selectively include certain pre-activity ceremonies or events. Normally the "play" should include a related series of activities that could potentially result in a victory by one contestant and a loss by the other contestant.

It is to be understood that the temporal bounds of a particular type of "play" does not necessarily start or end at a particular instance, but rather at a time generally coincident with the start and end of the play or otherwise based upon, at least in part, a time (e.g., event) based upon a play. A summarization of the video is created by including a plurality of video segments, where the summarization includes fewer frames than the original video from which the summarization was created. A summarization that includes a plurality of the plays of the event provides the viewer with a shorted video sequence while permitting the viewer to still enjoy the event because most of the exciting portions of the video are provided, preferably in the same temporally sequential manner as in the original video. In addition, it is to be understood that although summarization often achieves compression at the same time, it is different from video coding which aims at representing the original video with less data. In fact, summarization may be considered more concerned about the compact representation of the "content" in the video, whereas video coding is more concerned about representing the video signal itself as accurately and as bandwidth-efficient as possible.

Play Detection

Referring to FIG. 1A, a model of a class of sports video in terms of play is shown. The play portion is a basic segment of time during which an important action occurs in the game. The non-play is a segment of time during which a non-important action occurs in the game, or otherwise not determined to be a play. The inner loop illustrated in dashed lines indicates the possibility that two plays may occur consecutively or with a relatively short time period between the two plays.

Referring to FIG. 1B, a procedure for summarization of a video includes receiving a video sequence 20 that includes material to be summarized, where the content preferably includes at least a portion of a game or sporting event. Block 22 detects the start of a play of a video segment of a plurality of frames of the video. After detecting the start of the play, block 24 detects the end of the play, thereby defining a segment of video between the start of the play and the end of the play, namely, a "play". Block 26 then checks to see if the end of the video (or the portion to be processed) has been reached. If the end of the video has not been reached block 26 branches to block 22 to detect the next play. Alternatively, if the end of the video has been reached then block 26 branches to the summary description 28. The summary description defines those portions of the video sequence 20 that contain the relevant segments for the video summarization. The summary description may be compliant with the MPEG-7 Summary Description Scheme or TV-Anytime Segmentation Description Scheme. A compliant media browser may apply the summary description to the input video to provide summarized viewing of the input video without modifying it. Alternatively, the summary description may be used to edit the input video and create a separate video sequence. The summarized video sequence may comprise the selected segments which excludes at least a portion of the original video other than the plurality of segments. Preferably, the summarized video sequence excludes all portions of the original video other than the plurality of segments.

One component of the summarization procedure depicted in FIGS. 1A and 1B is the detection of an event, or "play." If the start and end points of all plays are detected, then the system may string all the plays together to obtain a summary from the original video and perform some post processing to smooth the transition boundaries, such as using dissolving techniques to reduce abrupt change between plays and smoothing the audio filed for better auditory effects. Further, the summary should ideally contain only those segments comprising a "play" as earlier defined (or portions of plays), thus providing a compact representation of the original tournament. With a compact representation the user can spend less time watching it while maintaining most of the excitement of the original game.

One of the difficulties in the detection of a "play" in a sporting broadcast is the detection of the events. However, the present inventors have come to the realization that for sporting broadcasts, and other broadcasts, the general video capturing and production patterns that have been adopted by virtually all of the broadcast companies permits the detection of the events. Hence, relatively low-level visual features may be used for event detection that are relatively invariant.

With the summarization being determined based upon low-level characteristics of the video, the system should detect an event (e.g., a play). In contrast to a generic summarization scheme which uses for example color histograms as the cue for key frame detection or scene classification, the different plays may contain colors which sweep a large range of color (in terms of histogram), yet all the frames belong to the same event, and may be used to form an uninterrupted video clip.

Football Play Detection

The present inventors then considered how to detect a "play" from a football video in a robust, efficient, and computationally effective manner. After extensive analysis of a typical football game it was determined that a football game is usually captured by cameras positioned at fixed locations around the football field, with each camera typically capable of panning, tilting, and zooming. Each play in a football game normally starts with the center hiking the ball, such as toward the quarterback or kicker. Further, a hiking scene, in which the center is about to hike the ball, is usually captured from a camera location to the side of the center. This camera angle is typically used because it is easier to observe the movements of all of the parties involved (the offense, the center, the quarterback, the receivers, the running back, and the defense) from this viewpoint. Thus a play typically starts with a frame such as shown in FIG. 2.

While an attempt to determine a hiking scene may include complex computationally intensive analysis of the frame(s) to detect the center, the quarterback, or the kicker, and the offense/defense, together with appropriate motion, this generally results in non-robust hiking scene detection. To overcome this limitation the present inventors were dumbfounded to recognize that the scenes used to capture a football video typically use the same set of camera angles. The football game normally includes cameras sitting either on one side of the field and on the two ends of the field. The side cameras are normally located in the stadium above the 25, 50, and 25 yard lines, and the two end cameras are located at the ends of the fields. There may be additional cameras, such as handheld cameras, but most of the events are captured by the side cameras and the end cameras. In general there are two different types of plays, namely, place kicks and regular plays (e.g., plays that are not place kicks). In general, place kicks (which include the kick-offs, extra point attempts, and field goal attempts) are usually captured by a camera near the end of the field, while a regular play (including runs, passes, and punts) is usually captured by a side camera. It is also noted that a kick-off is usually captured by an end camera followed by a side camera. Accordingly, the different plays of a football video may be categorized as one of two different types of plays, namely, a place kick, and a regular play.

The regular play typically starts with a frame such as that shown in FIG. 2. The camera then follows the ball until the ball is called dead, at which time the current regular play ends. After the end of the regular play there is typically a camera break, at which time the camera views other activity, such as the commentators or the fans. The time between the camera break and the start of the next play is usually not exciting and thus should not be included in the summary.

The place kick typically starts with a frame such as that shown in FIG. 3, and it normally ends with a camera break, in a manner similar to the regular play. For the place kick, there are normally more than one camera break before the end of the play, such as for example, a first camera break at the switch from the end camera to the side camera, and a second camera break when the play ends.

To determine a start of a play, such as those shown in FIGS. 2 and 3, the present inventors considered criteria that may be suitable to characterize such an event. The criteria to determine the start of the play is based on anticipated characteristics of the image, as opposed to analyzing the content of the video to determine the actual events. One criteria that may be used to determine the start of a play is the field color. Under the assumption that a camera provides a typical start frame like those shown in FIG. 2 or 3, it may be observed that the field has a generally green color. Accordingly, a characteristic of the start of a play may be if a sufficient spatial region of the frame has the generally green color. The sufficient spatial generally green region may be further defined by having shape characteristics, such as substantially straight edges, a set of substantially parallel edges, a four-sided polygon, etc. Further, the spatial region of the generally green color is preferably centrally located within the frame. Thus, it would initially appear that the start of a play can be detected by locating frames with a generally green dominant color in the central region. The aforementioned color test is useful in detecting the start of a play. However, after further analysis it was determined that merely detecting the generally green dominant color centrally located is sufficient but may be insufficient for a robust system. For example in some implementations, a dominant generally green color may be a necessary condition but not a sufficient condition for determining the start frame of play.

For example, the color characteristic of a central spatial generally green region may exist when the camera is focused on a single player on the field prior to a play. In addition, the precise color of the generally green color captured by the camera varies from field to field, from camera to camera, and from day to night. In fact, even for a given game, since it may start in late afternoon and last into early evening, the lighting condition may change, causing the generally green color of the same field to vary significantly during the video. Moreover, the generally green field color is typically not uniform and includes variations. Thus it is preferably not to use a narrow definition of the generally green color (e.g., excluding other non-green specific colors). Therefore, it is preferable to use a broad definition of generally green. If a broad definition of a generally green color is used, such as ones that includes portions of other colors, then a greater number of non-play scenes will be identified.

With the generally green color of the field not being constant, it is desirable to calibrate the generally green color for a specific football video. Further, it is desirable to calibrate the generally green color for a specific portion of a football video, with the generally green color being recalibrated for different portions of the football video. Referring to FIG. 4, using the hue component in the HSV color space as an example, the preferred system provides a range of generally green colors, such as G.sub.low and G.sub.high, with generally green being defined there between. The G.sub.low and/or G.sub.high may be automatically modified by the system to adapt to each particular football video and to different portions of the video.

With the variation of the field color even within a game, the present inventors determined that a color histogram H.sub.g of the generally green color in addition to a range given by G.sub.low and G.sub.high, provides a more accurate specification of the field color. The H.sub.g may calibrated for a specific football video. Also H.sub.g may be calibrated for a specific portion of the football video, with the H.sub.g being recalibrated for different portions of the football video. Even with two frames of the video showing the field the resulting color histograms will tend to be different. Thus, it is useful to estimate the extent to which the field color histograms vary in a particular football video, or portion thereof. It is preferable to use the field scenes, however detected, from which to estimate the color histograms.

The following technique may be used to determine G.sub.low, G.sub.high, and H.sub.g. Referring to FIG. 5, for all (or a portion of) the frames containing the field all the generally green pixels are located. For this initial determination preferably the generally green pixels are defined to include a large interval. The interval may be defined as G0=[G0.sub.low, G0.sub.high]. Next a statistic measure of the generally green pixels is calculated, such as the mean hue green value G.sub.mean of all the pixels. Next G.sub.low and G.sub.high may be set. One technique for setting G.sub.low and G.sub.high is: G.sub.low=G.sub.mean-g, G.sub.high=G.sub.mean+g, where g is a constant such that G.sub.high-G.sub.low<G0.sub.high-G0.sub.low. In essence, the technique narrows (i.e., reduces its gamut) the range of generally green colors based on color based information from the football video.

The following technique may be used to determine the color histogram H.sub.g. Referring to FIG. 6, all (or a portion of) the frames containing the field are selected. Within these field frames all (or a portion on) the pixels falling in the range of G.sub.low and G.sub.high are selected. Other ranges of generally green colors may likewise be used. The color histogram H.sub.i for each of these sets of pixels in each of the frames is then determined. Then H.sub.g is computed as a statistical measure, such as the average, of all the calculated color histograms H.sub.i. In particular the variation of H.sub.g may be calculated as follows:

For any frame containing the field, compute the error between H.sub.i and H.sub.g: e.sub.i=.parallel.H.sub.g-H.sub.i.parallel. where .parallel..parallel. is the L.sub.1 norm. The sample mean is computed as:

.times..times. ##EQU00001## The sample standard deviation of all the errors is calculated as:

.times. ##EQU00002##

with N being the number of frames, v being a measure for evaluating how a color histogram is different from the average H.sub.g.

With the green color being calibrated, the system may test if a frame is likely the start of a play by checking the following two conditions:

(1) if the frame has more than P.sub.1% generally green pixels; (2) if the color histogram H.sub.1 of these generally green pixels is close enough to H.sub.g. The first condition may be examined by counting the number of pixels whose hue value falls in G.sub.low, G.sub.high. The second condition may be examined by checking if the difference between H.sub.1 and H.sub.g is smaller than a threshold, i.e., if .parallel.H.sub.1-H.sub.g.parallel.<T.sub.h. The threshold T.sub.h may be determined as: T.sub.h=m.sub.e+cv, where c is a constant, typically 3 or 4. If both conditions are satisfied, then a potential start is detected, and this frame may then be further checked by other modules if it is desirable to confirm a detection. If however, the frame has only more than P.sub.2% green pixels (P.sub.2<P.sub.1), and the second condition is satisfied, then the field line detection module described later should be used to increase the confidence of an accurate determination of a potential start of a play.

After consideration of actual frames of the start of a play in football videos the present inventors observed that sometimes the start frames contain non-field regions on the top and the bottom, and further may contain editing bars on the side or on the bottom. These factors are not especially compatible with the use of the thresholds P.sub.1 and P.sub.2, as previously described. For the thresholds P.sub.1 and P.sub.2 to be more robust, only the center region (e.g., primarily generally within such non-field regions and editing bars) of the frame should be used when computing the percentages. Referring to FIG. 7, the center region may be defined as follows: (1) scan a frame row-by row, starting from the first row, until a row that has dominant generally green pixels is located, or until a predetermined maximum is reached, whichever occurs first; (2) scan the frame row-by-row, starting from the bottom row, until a row that has dominant generally green pixels is located, or until a predetermined maximum is reached, whichever occurs first; (3) scan the frame column-by-column, starting from the right column until a column that has dominant generally green pixels is located, or until a predetermined maximum is reached, whichever occurs first; (4) scan the frame column-by-column, starting from the left column until a column that has dominant generally green pixels is located, or until a predetermined maximum is reached, whichever occurs first; (5) the locations at which the scanning stopped (e.g., found the dominant generally green color or otherwise a predetermined maximum), defines the central region of the frame. The preferred predetermined maximums are 1/4 of the row number as the constant in the scanning of the rows and 1/6 of the column number as the constant in the scanning of the columns.

After further consideration of the football video, the present inventors likewise observed a pattern exhibited by the football video at the start of a play, namely, the field lines. The presence of the field lines is a strong indication of the existence of a corresponding field being viewed by the camera. The field lines may be characterized by multiple substantially parallel spaced apart substantially straight lines or lines on a contrasting background. The field lines may alternatively be characterized by multiple spaced apart generally white lines. In addition, the field lines may be characterized as a pattern of lines on a background primarily a generally green color. Also, the field lines may be further constrained as being of a sufficient length relative to the size of the field or image. In the preferred system, the field lines are characterized as two, three, or four of the above. This length consideration removes shorter lines from erroneously indicating a field. The identification of the frames of video representing fields using the field lines may be used as the basis for the color calibration, if desired.

Referring to FIG. 8, the preferred system includes candidate frame selection by using an initial green specification, such as G0=[G0.sub.low, G0.sub.high]. Then those frames with a primary color G0 are identified. A green mask may be obtained by setting a value of "1" to locations defined by the G0 color and "0" to the other locations. The green mask may then be diluted, if desired, to allow the inclusion of small regions adjacent to the green G0 region. The edge detection may then be performed on the frames followed by filtering with the green mask. This step is intended to eliminate those edge pixels that are not on the generally green background. A line detection is then performed on the filtered edge map, such as with a Hough transform, to get lines that are longer than L.sub.min. It is to be understood that any suitable technique may be used to identify the lines, and in particular the lines within a generally green background.

After experimentation with the line detection scheme there remains a small probability that such line detection will result in false positives, even in a generally green background. The present inventors further considered that an image of a field from a single viewpoint results in some distortion of the parallel alignment of the field lines. In particular, a plurality of the field lines will appear to converge at some point (or points). Preferably, all of the field lines will appear to pass through approximately the same disappearing point since the field lines are parallel to one another on the field. Referring to FIG. 9, a sample frame is shown. Referring to FIG. 10, the result of the edge detection is shown. Referring to FIG. 11, the parametric lines along the vertical direction are illustrated, with the lines passing generally through the same point.

In the preferred system, the condition that is used is detecting at least three lines that pass through approximately the same point when projected. This additional condition, especially when used in conjunction with previous field line determination, significantly decreases the likelihood of false positives. Similarly, when the frame is from an end camera, such as shown in FIG. 3, the field lines would appear to be nearly horizontal and parallel to each other in the image domain, which is likewise a test for determination of a field. As shown in FIG. 8, in either case (side view of the field or end view of the field) the task is to test if the lines are parallel in the physical world, and this is referred to as the parallelism test. After the parallelism test the green may be calibrated and the start of a play may be determined based upon these characteristics.

The present inventors observed that there are some cases where the field may contain multiple regions of clay which is of generally brown color. The color calibration technique described above can be similarly applied to deal with these cases so that the system can handle fields of generally green color, fields of generally green and generally brown colors, and fields of generally brown color. Other techniques may likewise be applied to the generally brown, or generally brown and generally green.

The present inventors observed that in many cases the two teams are lined up and most of the motion stops before the start of a play. At this point, the camera motion may tend to zoom in to get an improved picture and stays focused on the players until the play starts. Thus at the moment right before a play starts, there will tend to be no significant motion in the image domain (neither camera-induced motion nor player motion). Therefore, the present inventors determined that the camera motion may be used as an additional indicia of the start of a play. In many instances, a start-of-play will induce a zooming in camera motion that then stops zooming with the scene being free from significant motion. This is another characteristic that may be used to indicate the start of plays. This technique may likewise be used in conjunction with other techniques to decrease false positives.

There are several techniques that may be used for estimating camera motion. Some methods such as optical flow estimation may provide dense motion fields and hence provide relatively accurate motion estimation results. However, optical flow techniques and similar techniques, are computationally expensive. A less computationally expensive technique is to infer the camera motion from block-based motion compensation. In addition, the motion information is available without additional computation if the system is operating on compressed streams of encoded video, such as a MPEG-like bitstream. It has been determined that the translational motion can be accurately estimated from the motion vectors whereas zooming is not accurately estimated from the motion vectors. The inaccuracy of the motion vectors for zooming may be based on the varying rate of zooming and the scale changes induced by zooming. Therefore, the motion information is preferably used in the following manner: if the camera motion is not primarily translational, the system waits additional frames to confirm the start of a play; otherwise, the start-of-play is declared as long as other conditions are satisfied. A waiting period in the first has dual functions: firstly, it excludes from the summary some frames when the camera is zooming before a start of the play; and secondly, it makes the detection of the start-of-play more robust since more frames have been used to confirm the detection. FIG. 12 illustrates an example of computed motion vectors, when the camera is switched on after a play has started. It is not difficult to deduce that the camera is panning in this situation, based on the primary direction of the motion vectors. In this case a start-of-play may be declared.

As illustrated in FIGS. 2 and 3, in a start-of-play frame, the players appear as scattered blobs in the image. The blobs may be represented by their color and/or texture, and compared against a model of the anticipated color and/or texture for a player. The color and/or texture may be varied, based on the particular team's clothing. In this manner, the system is customizable for particular teams. In the case that there are scattered non-generally green blobs their color characteristics may be compared against a model. In addition, the system may determine, using other techniques, to determine potential start of play frames and use these frames as the basis to calculate color histograms for the players.

Referring to FIG. 13, at the start of the football play the each of the teams tend to line up in some manner. This line up of the players may be used as a characteristic upon which to determine the start of a play. The characteristic of a suitable line up of players includes a generally aligned set of non-generally green blobs (e.g., regions), such as the green mask shown in FIG. 14, as previously described. Further, the blobs should have a relatively small size, especially in relation to the size of the field. In contrast, a relatively large non-generally green blob, such as the green mask shown in FIG. 15, is more likely indicative of a close up of a player, such as shown in FIG. 16. To characterize the spatial distribution of the non-generally green regions the green masks may be projected into x and y directions, such as shown in FIG. 17 and FIG. 18. A high and wide peak in the projection, as shown in FIG. 18, is less likely to indicate the start of a play than a generally low set of peaks, as shown in FIG. 17. Another approach for analyzing the line up of players may be determining two distinctive groups of blobs lining up along both sides of a "line" that is parallel to the field lines.

After further consideration, the present inventors determined that if a hiking scene and accordingly a play segment is identified after locating only one candidate frame, then the system may be susceptible to false positives. By examining a set of consecutive frames (or other temporally related frames) and accumulating evidence, the system can reduce the false positive rate. Referring to FIG. 19, the following approach may be used to achieve temporal evidence of accumulation: when detecting a hiking scene, a sliding window of width w is used (e.g., w frames are considered at the same time). A hiking scene is declared only if more than p out of the w frames in the current window are determined to be hiking scene candidates, as previously described. A suitable value of p is such that p/w=70%. Other statistical measures may be used of a fixed number of frames or dynamic number of frames to more accurately determine hiking scenes.

To define the "generally green" color any color space may be used. The preferred color space is the HSV color space because it may be used without excessive computational complexity. Alternatively, a YUV color space may be used as shown in FIG. 20.

While the start of a "play" may be defined as a hiking scene the end of a play, according to the rules of football, can end in a variety of different ways. Image analysis techniques may be used to analyze the image content of the frames after a hiking frame to attempt to determine what occurred. Unfortunately, with the nearly endless possibilities and the difficultly of interpreting the content of the frames, this technique is at least, extremely difficult and computationally intensive. In contrast to attempting to analyze the content of the subsequent frames of a potential play, the present inventors determined that a more efficient manner for the determination of the extent of a play in football is to base the end of the play on camera activities. After analysis of a football video the present inventors were surprised to determine that the approximate end of a play may be modeled by scene changes, normally as a result of switching to a different camera or a different camera angle. The different camera or different camera angle may be modeled by determining the amount of change between the current frame (or set of frames) to the next frame (or set of frames).

Referring to FIG. 21A, a model of the amount of change between frames using a color histogram difference technique for an exemplary 1,000 frame video football clip is shown. The peaks typically correspond to scene cuts. The system may detect an end of play at around frame 649 by thresholding the color histogram difference. A gradual transition occurs around frame 350.

As previously noted the scene cuts may be detected by thresholding the color histogram differences. The selection of the an appropriate threshold level to determine