Tide filtering for video surveillance system

• US 7,424,167 B1
• Filed: 10/01/2004
• Issued: 09/09/2008
• Est. Priority Date: 10/01/2004
• Status: Active Grant

• Current Assignee: Motorola Solutions, Inc.
• Original Assignee: ObjectVideo Incorporated (Alarm.com Holdings, Inc.)
• Inventors: Chosak, Andrew J., Brewer, Paul C., Lipton, Alan J., Clark, John I. W., Venetianer, Péter L.
• Primary Examiner(s): Johns; Andrew W.
• Assistant Examiner(s): Smith; Jeffrey S
• Application Number:
  US10/954,479
• Time in Patent Office: 1,439 Days
• Field of Search: 382/263, 382/266, 382/271
• US Class Current: 382/263
• CPC Class Codes: G06V 20/52 Surveillance or monitoring of activities, e.g. for recognising suspicious objects recognising microscopic objects G06V20/69

Abstract

A machine-accessible medium containing software code that, when read by a computer, causes the computer to perform method for filtering a video in a video system. The method may include generating a foreground mask and a background model from a video, where the foreground mask comprises moving pixels in the video and the background model comprises a statistical description, including a mean and a variance value, for each pixel in the video. Then the background model variance is filtered with a one-dimensional high pass filter in a single orientation. Next, for each linear grouping of pixels in the single orientation of the filtered variance, a first edge pixel is detected between a high-variance group of pixels and a low-variance group of pixels. Then a first waterline position is detected as an area bounded by the first edge pixels.

Claims

1. A machine-accessible medium containing software code that, when read by a computer, causes the computer to perform a method for filtering a video in a video system, comprising:

generating a foreground mask and a background model from a video, wherein said foreground mask comprises moving pixels in the video and said background model comprises a statistical description, including a mean and a variance value, for each pixel in the video;

filtering said background model variance with a one-dimensional high pass filter in a single orientation;

for each linear grouping of pixels in said single orientation of said filtered variance, detecting a first edge pixel between a high-variance group of pixels and a low-variance group of pixels; and

detecting a first waterline position as an area bounded by said first edge pixels.

2. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

prior to filtering, determining a direction in which water in the video enters a frame of the video; and

orienting the video such that said direction is substantially aligned with said single orientation.

3. The machine-accessible medium of claim 2, wherein said determining a direction comprises at least one of:

automatically detecting said direction;

or

receiving user input designating said direction.

4. The machine-accessible medium of claim 3, wherein said automatic detecting comprises:

monitoring said background model around boundaries of said frame over time; and

if the variance values at one of said boundaries of said frame are consistently higher than at other said boundaries, choosing said one of said boundaries as said direction.

5. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

sub-sampling the video prior to said filtering.

6. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

correcting said first waterline position when no waterline is detected due to water covering at least one complete linear grouping of pixels in said single orientation of the video.

7. The machine-accessible medium of claim 6, wherein said correcting comprises:

identifying at least one complete linear grouping of pixels in said single orientation where no waterline is detected;

fitting a line to said first waterline; and

determining whether said fitted line suggests that said at least one complete linear grouping of pixels in said single orientation is covered by water.

8. The machine-accessible medium of claim 7, wherein said determining comprises:

determining a first representative measure of the variance of pixels on a first side of said edge pixels and a second representative measure of the variance of pixels on a second side of said edge pixels, wherein said first side is a known water area and said second side is a known land area;

determining a third representative measure of the variance of pixels in said at least one complete linear grouping of pixels in said single orientation;

comparing said third representative measure of the variance to said first representative measure of the variance and said second representative measure of the variance; and

identifying said at least one complete linear grouping of pixels in said single orientation as being water when said third representative measure of the variance is more similar to said first representative measure of the variance than said second representative measure of the variance.

9. The machine-accessible medium of claim 7, wherein said determining comprises:

generating a first intensity histogram for a first side of said edge pixels and a second intensity histogram for a second side of said edge pixels, wherein said first side is a known water area and said second side is a known land area;

generating a third intensity histogram for said at least one complete linear grouping of pixels in said single orientation;

comparing a peak of said third intensity histogram to a peak of said first intensity histogram and a peak of said second intensity histogram; and

identifying said at least one complete linear grouping of pixels in said single orientation as being water when said peak of said third intensity histogram is more similar to said peak of said first intensity histogram than to said peak of said second intensity histogram.

10. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

correcting said waterline position to remove an object included in a water area bounded by said detected waterline.

11. The machine-accessible medium of claim 10, wherein said correcting comprises:

fitting a line to said detected waterline;

detecting an object area wherein said fitted line and said detected waterline are substantially not in alignment;

comparing a background variance of said detected object area to a variance of a known land area and a variance of a known water area in the video;

identifying said detected object area as not being water when said variance of said detected object area is more similar to the variance of said known land area;

repeating said edge pixel detection for each linear grouping of pixels in said single orientation containing said detected object area to find a second edge pixel for each said row; and

correcting said detected waterline with said second edge pixels.

12. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

spatial and temporal filtering to smooth said detected waterline.

13. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

using said detected waterline to remove pixels corresponding to water from said foreground mask.

14. The machine-accessible medium of claim 1, wherein the software code causes the computer to perform the method further comprising:

detecting a swash zone having a leading edge and a trailing edge; and

excluding said swash zone from said foreground mask.

15. The machine-accessible medium of claim 14, wherein said detecting a swash zone further comprises:

for each linear grouping of pixels in said single orientation of said filtered variance, detecting a second edge pixel between a low-variance group of pixels and a high-variance group of pixels;

detecting a second waterline position as an area bounded by said second edge pixels, wherein said swash zone lies between said first and said second waterlines.

16. A system for filtering in a video system, comprising:

a background segmentation module adapted to generate a foreground mask and a background model from a video, wherein said foreground mask comprises moving pixels in the video and said background model comprises a statistical description, including a mean and a variance value, for each pixel in the video; and

a tide-detection module adapted to filter said background model variance with a one-dimensional high pass filter in a single orientation, to detect for each linear grouping of pixels in said single orientation in said filtered variance a first edge pixel between a high-variance group of pixels and a low-variance group of pixels, and to detect a first waterline position as an area bounded by said first edge pixels.

17. The system of claim 16, further comprising:

a pre-processing module adapted to perform at least one of;

sub-sampling the video;

or

determining a direction in which water in the video is moving and orienting the video such that the water is moving substantially in said single orientation.

18. The system of claim 16, further comprising:

a post-processing module adapted to perform at least one of;

correcting said waterline position to remove an object included in a water area bounded by said detected waterline;

or

identifying at least one complete linear grouping of pixels in said single orientation of the video as being water.

19. The system of claim 16, further comprising:

a spatial and temporal filtering module adapted to smooth said detected waterline.

20. The system of claim 16, further comprising:

a masking module adapted to use said detected waterline to remove pixels corresponding to water from said foreground mask.

21. The system of claim 20, wherein said tide-detection module is further adapted to detect a swash zone having a leading edge and a trailing edge;

and wherein said masking module is adapted to exclude said swash zone from said foreground mask.

22. A method for filtering a video in a video system, comprising:

generating a foreground mask and a background model from a video, wherein said foreground mask comprises moving pixels in the video and said background model comprises a statistical description, including a mean and a variance value, for each pixel in the video;

filtering said background model variance with a one-dimensional high pass filter in a single orientation;

for each linear grouping of pixels in said single orientation of said filtered variance, detecting a first edge pixel between a high-variance group of pixels and a low-variance group of pixels; and

detecting a first waterline position as an area bounded by said first edge pixels.

23. The method of claim 22, further comprising:

prior to filtering, determining a direction in which water in the video enters a frame of the video; and

orienting the video such that said direction is substantially aligned with said single orientation.

24. The method of claim 22, further comprising:

sub-sampling the video prior to said filtering.

25. The method of claim 22, further comprising:

correcting said first waterline position when no waterline is detected due to water covering at least one complete linear grouping of pixels in said single orientation of the video.

26. The method of claim 22, further comprising:

spatial and temporal filtering to smooth said detected waterline.

27. The method of claim 22, further comprising:

using said detected waterline to remove pixels corresponding to water from said foreground mask.

28. The method of claim 22, further comprising:

detecting a swash zone having a leading edge and a trailing edge; and

excluding said swash zone from said foreground mask.