Method for detecting water regions in video

1. A computer-based method for automatic detection of water regions in a video, the method comprising:

generating a water map from the video; and

outputting the water map to an output medium,

wherein a computer performs generating the water map from the video, and

wherein generating the water map comprises:

a. estimating a water map of the video from video frames,

wherein estimating the water map of the video comprises:

i. storing a first frame of the video in memory as a reference frame;

ii. initializing a variance map and the initial water map;

iii. retrieving a next frame of the video and dividing the next frame into a grid of at least two blocks;

iv. calculating a normalized correlation between the reference frame and each block in the next frame;

v. calculating a gray scale gradient at each pixel in the next frame;

vi. calculating the variance for each block in the grid by adding the variance of each pixel in the block;

vii. repeating (iii) through (vi) for a specified number of next frames;

viii. calculating a multiplicative product of the normalized correlation, the gray scale gradient, and the variance;

ix. comparing the product with an average value of the multiplicative product over all blocks;

x. marking a block as a water region in the initial water map when the product is greater than the average, and marking the block as a non-water region when the product is not greater than the average; and

xi. repeating (viii) through (x) for each block, whereby the initial water map is created.

2. The method of claim 1 , further comprising:

b. estimating a statistical water model from the estimated water map;

c. re-classifying the estimated water map using the water model and frames of the video; and

d. refining the re-classified water map.

3. The method of claim 2 , wherein the water model is a histogram comprising a plurality of bins and wherein estimating the statistical water model further comprises:

i. for each pixel in the initial water map:

(a) determining whether the pixel is both a water pixel and not a foreground object pixel;

(b) determining a corresponding bin for the pixel, when the pixel is both a water pixel and not a foreground pixel;

(c) increasing a likelihood that the corresponding bin, and bins neighboring the corresponding bin, are water; and

(d) repeating (a) through (c) for all pixels in the initial water map, whereby the water model is complete; and

ii. computing an adaptive threshold based on an average of all bins in the histogram.

4. The method of claim 3 , wherein step (ii) of computing an adaptive threshold comprises: computing an average of all the plurality of bins.

5. The method of claim 4 , wherein re-classifying the estimated water map further comprises:

i. initializing a water region map corresponding to a frame of the video;

ii. for each pixel in the frame:

(a) identifying the bin corresponding to the pixel;

(b) comparing the value of the corresponding bin to the adaptive threshold;

(c) marking the pixel as a definite water region when the value of the bin is greater than the adaptive threshold; and

(d) marking the pixel as a non-water region when the value of the bin is not greater than the adaptive threshold.

6. The method of claim 2 , wherein refining the re-classified water map comprises at least one of eroding the water map or dilating the water map.

7. The method of claim 1 , further comprising:

detecting an object in the video sequence based on the water map.

8. The method of claim 1 , further comprising:

tracking an object in the video sequence based on the water map.

9. The method of claim 1 , further comprising:

classifying an object in the video sequence based on the water map.

10. The method of claim 1 , further comprising:

detecting an event in the video sequence based on the water map.

11. A non-transitory computer-readable medium containing instructions that, when executed by a processor, cause the processor to perform the method according to claim 1 .

12. A video processing system comprising:

a computer system; and

the non-transitory computer-readable medium according to claim 11 .

13. A video surveillance system comprising:

at least one camera to generate an input video sequence; and

the video processing system according to claim 12 .

14. A method comprising:

one of downloading software using a computer to the computer over a network or providing downloadable software from a computer to another computer over a network, which software when executed by the computer, causes the computer to perform operations comprising the method of claim 1 .

15. A computer system for detecting water regions in a video, wherein the computer system comprises a non-transitory computer-readable medium comprising software modules for video processing for detecting water regions in the video, wherein the software modules comprise:

a water detector module to receive a video to detect the presence of water in the video; and

means for outputting the water map to an output medium,

wherein the water detector module further comprises a water estimation module to receive the video and to create an estimated water map, and

wherein the computer system further comprises:

a water model training module, to create a water model using the estimated water map;

a re-classification module to re-classify the estimated water map using the water model and frames of the video, and to create a water map; and

a refinement module to refine the water map.

16. The computer system according to claim 15 , wherein said event specification interface comprises a graphical user interface.

17. The computer system of claim 15 , wherein the detection module provides feedback to the water detector module.

18. The computer system of claim 15 , further comprising:

a video analysis system to receive the video and the water map.

19. The computer system of claim 15 , further comprising at least one of:

a detection module to detect an object in the video;

a tracking module to track a detected object;

a classification module to classify a detected object;

an event detection module to receive an output of said water detector module and to output one or more detected events; or

an event specification interface to provide one or more events of interest to said event detection module.

20. An apparatus for water detection, wherein the apparatus comprises a a non-transitory computer-readable medium comprising software modules for video processing for water detection, which when executed by a computer, causes the computer to perform the operations of:

generating a water map from the video; and

outputting the water map to an output medium,

wherein the operations comprising generating the water map comprise:

a. estimating a water map of the video,

b. estimating a statistical water model from the estimated water map,

c. re-classifying the estimated water map using the water model and frames of the video; and

d. refining the re-classified water map.

21. The apparatus of claim 20 , wherein the apparatus comprises:

application-specific hardware to emulate a computer and/or software, the application-specific hardware to perform the operations of claim 20 .

22. An apparatus as in claim 20 , wherein the apparatus is embodied in a video camera.

23. The apparatus of claim 20 , wherein estimating the water map of the video comprises:

i. storing a first frame of the video in memory as a reference frame;

ii. initializing a variance map and the initial water map;

iii. retrieving a next frame of the video and dividing the next frame into a grid of at least two blocks;

iv. calculating a normalized correlation between the reference frame and each block in the next frame;

v. calculating a gray scale gradient at each pixel in the next frame;

vi. calculating the variance for each block in the grid by adding the variance of each pixel in the block;

vii. repeating (iii) through (vi) for a specified number of next frames;

viii. calculating a multiplicative product of the normalized correlation, the gray scale gradient, and the variance;

ix. comparing the product with an average value of the multiplicative product over all blocks;

x. marking a block as a water region in the initial water map when the product is greater than the average, and marking the block as a non-water region when the product is not greater than the average; and

xi. repeating (viii) through (x) for each block, whereby the initial water map is created.

24. The apparatus of claim 20 , wherein the water model is a histogram comprising a plurality of bins and wherein estimating the statistical water model further comprises:

i. for each pixel in the initial water map:

(a) determining whether the pixel is both a water pixel and not a foreground object pixel;

(b) determining a corresponding bin for the pixel, when the pixel is both a water pixel and not a foreground pixel;

(c) increasing a likelihood that the corresponding bin, and bins neighboring the corresponding bin, are water; and

(d) repeating (a) through (c) for all pixels in the initial water map, whereby the water model is complete; and

ii. computing an adaptive threshold based on an average of all bins in the histogram.

25. The apparatus of claim 24 , wherein step (ii) of computing an adaptive threshold comprises: computing an average of all the plurality of bins.

26. The apparatus of claim 20 , wherein re-classifying the estimated water map further comprises:

i. initializing a water region map corresponding to a frame of the video;

ii. for each pixel in the frame:

(a) identifying the bin corresponding to the pixel;

(b) comparing the value of the corresponding bin to the adaptive threshold;

(c) marking the pixel as a definite water region when the value of the bin is greater than the adaptive threshold; and

(d) marking the pixel as a non-water region when the value of the bin is not greater than the adaptive threshold.

27. The apparatus of claim 20 , wherein refining the re-classified water map comprises at least one of eroding the water map or dilating the water map.

28. A video processing system comprising:

a video processing device to accept an input video and detect a water region in the input video; and

an output medium to output information regarding the detected water region,

wherein the video processing device further comprises:

a water estimation module to receive the video and to create an estimated water map;

a water model training module, to create a water model using the estimated water map;

a re-classification module to re-classify the estimated water map using the water model and frames of the video, and to create a water map; and

a refinement module to refine the water map.