Video demultiplexing based on meaningful modes extraction

• US 7,328,198 B1
• Filed: 12/31/2004
• Issued: 02/05/2008
• Est. Priority Date: 12/31/2004
• Status: Active Grant

• Current Assignee: Cognitech, Inc.
• Original Assignee: Cognitech, Inc.
• Inventors: Yu, Ping, Rudin, Leonid I., Morel, Jean Michel, Lisani, Jose Luis
• Primary Examiner(s): Starks, Jr.; Wilbert L
• Application Number:
  US11/028,470
• Time in Patent Office: 1,131 Days
• Field of Search: 706/20, 706/45, 382/156, 382/157, 382/224, 345/473, 370/252, 370/352
• US Class Current: 706/20
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06V 10/40 Extraction of image or video features
• CPC Class Codes: G06V 20/41 Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items segmenting video sequences G06V20/49

Abstract

Methods and systems for grouping video frames. The video frames may be from a multiplexed stream of video frames from a plurality of video sources. In some embodiments video frames are grouped by determining if the video frames share meaningful modes for values representative of features of the video frames.

Claims

1. A method, using a computer, of grouping video images, the video images being from a plurality of source cameras providing images of locations, the method comprising:

calculating values of a plurality of video image features for a plurality of video images;

determining meaningful concentrations of the values; and

grouping video images having common values of features in the concentrations so as to group the video images by source camera.

2. The method of claim 1 wherein the meaningful concentration is a concentration greater than an average expectation concentration.

3. The method of claim 1 wherein the meaningful concentration is determined using at least one of a statistical distribution, a stochastic process, a uniform distribution, a Gaussian distribution, or a binomial distribution.

4. The method of claim 1 wherein the concentration of values defines a meaningful mode in a histogram.

5. The method of claim 1 wherein determining a meaningful concentration of values of comprises:

assigning to each value a number of occurrences of the value, the number of occurrences corresponding to the number of video images having the value of the feature;

comparing the number of occurrences to an expected concentration; and

determining intervals of values above the expected concentration.

6. The method of claim 5 wherein grouping video images having common values of features in the concentrations comprises:

identifying video images having feature values in the same meaningful modes.

7. A method, using a computer, of assigning video images to sub-sets, the video images being from a plurality of cameras, comprising:

receiving M video images;

selecting n features characterizing at least part of video images of the M video images;

determining a value of a first feature of the n features for each of the M video images;

determining meaningful modes for the values; and

assigning video images of the M video images having values in the same meaningful modes to the same sub-sets so that each sub-set includes images from less than all of the plurality of cameras.

8. The method of claim 7 further comprising:

iteratively;

determining a value of an ith feature of the n features for video images in a sub-set;

determining meaningful modes for the values; and

assigning video images of the sub-set having values in the same meaningful modes to further sub-sets so as to reduce the number of cameras which sourced images in the further sub-sets.

9. The method of claim 8 further comprising identifying meaningful modes of the further sub-sets.

10. The method of claim 9 further comprising receiving a further plurality of video images.

11. The method of claim 10 further comprising determining feature values for the n features for the further plurality of video images.

12. The method of claim 11 further comprising determining video images of the further plurality of video images having values corresponding to meaningful modes of the further sub-sets.

13. The method of claim 12 further comprising assigning the further plurality of video images having values corresponding to the same meaningful modes of the further sub-sets to the further sub-sets.

14. The method of claim 7 wherein the video images are comprised of pixels, each pixel defined by pixel values.

15. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by a function of the pixel values for at least some of the pixels.

16. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by a function of the pixel values for all of the pixels.

17. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by an average gray level of the video image.

18. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by an average color level of the video image.

19. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by pixels with an attribute above a level of the attribute.

20. The method of claim 14 wherein at least some of the n features characterizing at least part of video images have values determined by pixels with an attribute below a level of the attribute.

21. A method, using a computer, of demultiplexing a multiplexed video stream having video frames provided by a plurality of video sources producing location images, the method comprising:

iteratively;

receiving a video frame;

computing a plurality of video image feature values for the video frame;

comparing the plurality of video image feature values for the video frame with a plurality of video image feature values of a plurality of reference video frames;

assigning the video frame to a set corresponding to a specific reference video frame of the plurality of reference video frames if differences between the plurality of video image feature values of the video frame and the plurality of video image feature values of the specific reference video frame are within a meaningful mode about zero, thereby indicating that the video frame and the specific reference video frame derive from the same video source of the plurality of video sources; and

identifying the video frame as a reference video frame from among the video frames provided by the plurality of video sources if differences between the plurality of video image feature values of the video frame and the plurality of video image feature values of the plurality of reference video frames are not within a meaningful mode about zero.

22. The method of claim 21 wherein each of the video image feature values are a function of the entire video frame.

23. The method of claim 22 wherein the differences between video image feature values are within a meaningful mode about zero if all of the differences between video image feature values are within a meaningful mode about zero.

24. The method of claim 21 wherein each of the video image feature values are a function of only a part of the video frame.

25. The method of claim 24 wherein the differences between video image feature values are within a meaningful mode about zero if a substantial amount of the differences between video image feature values are within a meaningful mode about zero.