Foreground detector for video analytics system
Techniques are disclosed for creating a background model of a scene using both a pixel based approach and a context based approach. The combined approach provides an effective technique for segmenting scene foreground from background in frames of a video stream. Further, this approach can scale to process large numbers of camera feeds simultaneously, e.g., using parallel processing architectures, while still generating an accurate background model. Further, using both a pixel based approach and context based approach ensures that the video analytics system can effectively and efficiently respond to changes in a scene, without overly increasing computational complexity. In addition, techniques are disclosed for updating the background model, from frame-to-frame, by absorbing foreground pixels into the background model via an absorption window, and dynamically updating background/foreground thresholds.
1. A computer-implemented method, comprising:
receiving image data, wherein the image data includes one or more appearance values for each of a plurality of pixels;
creating a candidate background image and a candidate foreground image by comparing the one or more appearance values of each pixel to a background model of the image data; and
performing one or more context based evaluations on one or more of the pixels classified as depicting foreground, wherein the context based evaluations selectively reclassifies one or more of the pixels as depicting foreground or background based on the classification of other pixels in the image data as depicting either foreground or background.
2. The method of claim 1 , wherein comparing the one or more appearance values of each pixel to a background model of the image data comprises:
for each pixel of the plurality of pixels:
determining a distance between the pixel and a corresponding pixel in the background model,
upon determining the distance exceeds a threshold, classifying the pixel as depicting foreground, and
upon determining the distance does not exceed the threshold, classifying the pixel as depicting background.
3. The method of claim 2 , wherein the threshold is dynamically determined for each pixel based on a camera noise model.
4. The method of claim 2 , wherein the background model provides, for each pixel in the image data, a distribution modeling each of the one or more appearance values for the given pixel in the image data and a respective mean and variance of each distribution.
5. The method of claim 1 , wherein one of the context based evaluations comprises: dilating one or more of the pixels classified as depicting foreground and reclassifying, as foreground, one or more of the pixels classified as background with overlap with a specified number of the dilated pixels.
6. The method of claim 1 , wherein one of the context based evaluations comprises eroding one or more contiguous groups of foreground pixels by reclassifying the pixels in the one or more contiguous groups as background.
7. The method of claim 1 , wherein one of the context based evaluations comprises:
identifying a contiguous region of pixels classified as foreground;
determining a normalized cross correlation score between the contiguous region of pixels in the video game classified as foreground and a corresponding region of pixels in a background image of the background model; and
upon determining the normalized cross correlation score satisfies a specified threshold, reclassifying the pixels in the contiguous region of pixels as depicting background.
8. A non-transitory computer-readable storage medium containing a program, which when executed on a processor, performs an operation for generating a background model of image data, the operation comprising:
receiving image data, wherein the image data includes one or more appearance values for each of a plurality of pixels;
classifying each pixel as depicting either foreground or background by comparing the one or more appearance values of each pixel to a background model of the image data; and
performing one or more context based evaluations on one or more of the pixels classified as depicting foreground, wherein the context based evaluations selectively reclassifies one or more of the pixels as depicting foreground or background based on the classification of other pixels in the image data as depicting either foreground or background.
9. The non-transitory computer-readable storage medium of claim 8 , wherein comparing the one or more appearance values of each pixel to a background model of the image data comprises:
for each pixel of the plurality of pixels:
determining a distance between the pixel and a corresponding pixel in the background model,
upon determining the distance exceeds a threshold, classifying the pixel as depicting foreground, and
upon determining the distance does not exceed the threshold, classifying the pixel as depicting background.
10. The non-transitory computer-readable storage medium of claim 9 , wherein the distance is a Mahalanobis distance.
11. The non-transitory computer-readable storage medium of claim 8 , wherein one of the context based evaluations comprises: dilating one or more of the pixels classified as depicting foreground and reclassifying, as foreground, one or more of the pixels classified as background with overlap with a specified number of the dilated pixels.
12. The non-transitory computer-readable storage medium of claim 8 , wherein one of the context based evaluations comprises eroding one or more contiguous groups of foreground pixels by reclassifying the pixels in the one or more contiguous groups as background.
13. The non-transitory computer-readable storage medium of claim 8 , wherein one of the context based evaluations comprises:
identifying a contiguous region of pixels classified as foreground;
determining a normalized cross correlation score between the contiguous region of pixels in the video game classified as foreground and a corresponding region of pixels in a background image of the background model; and
upon determining the normalized cross correlation score satisfies a specified threshold, reclassifying the pixels in the contiguous region of pixels as depicting background.
14. A system, comprising:
a video input source configured to provide image data;
a central processing unit (CPU);
a graphics processing unit (GPU); and
a memory containing a program, which, when executed on the CPU and GPU is configured to perform an operation for generating a background model of image data, the operation comprising:
receiving image data, wherein the image data includes one or more appearance values for each of a plurality of pixels,
classifying each pixel as depicting either foreground or background by comparing the one or more appearance values of each pixel to a background model of the image data, and
performing one or more context based evaluations on one or more of the pixels classified as depicting foreground, wherein the context based evaluations selectively reclassifies one or more of the pixels as depicting foreground or background based on the classification of other pixels in the image data as depicting either foreground or background.
15. The system medium of claim 14 , wherein comparing the one or more appearance values of each pixel to a background model of the image data comprises:
for each pixel of the plurality of pixels:
determining a distance between the pixel and a corresponding pixel in the background model,
upon determining the distance exceeds a threshold, classifying the pixel as depicting foreground, and
upon determining the distance does not exceed the threshold, classifying the pixel as depicting background.
16. The system of claim 15 , wherein the threshold is dynamically determined for each pixel based on a camera noise model.
17. The system of claim 15 , wherein the background model provides, for each pixel in the image data, a distribution modeling each of the one or more appearance values for the given pixel in the image data and a respective mean and variance of each distribution.
18. The system of claim 14 , wherein one of the context based evaluations comprises: dilating one or more of the pixels classified as depicting foreground and reclassifying, as foreground, one or more of the pixels classified as background with overlap with a specified number of the dilated pixels.
19. The system of claim 14 , wherein one of the context based evaluations comprises eroding one or more contiguous groups of foreground pixels by reclassifying the pixels in the one or more contiguous groups as background.
20. The system of claim 14 , wherein one of the context based evaluations comprises:
identifying a contiguous region of pixels classified as foreground;
determining a normalized cross correlation score between the contiguous region of pixels in the video game classified as foreground and a corresponding region of pixels in a background image of the background model; and
upon determining the normalized cross correlation score satisfies a specified threshold, reclassifying the pixels in the contiguous region of pixels as depicting background.