System and method for robust motion detection
View Patent ↗Method and system for detecting objects of interest in a camera monitored area are disclosed. Statistical analysis of block feature data, particularly Sobel edge and spatial high frequency responses is used to model the background of the scene and to segregate foreground objects from the background. This technique provides a robust motion detection scheme prone to catching genuine motions and immune against false alarms.
1. A method for detecting salient objects in a sequence of video frames captured from a scene, each frame comprising a plurality of pixels grouped in a plurality of image blocks, said method comprising:
(a) receiving pixel data and block feature data for each of the plurality of image blocks, the block feature data being at least one of Sobel edge and spatial high frequency response values for each pixel averaged over the block;
(b) classifying the blocks as background or foreground in a current frame based on temporal profile of the block feature data;
(c) identifying objects as a plurality of blobs, wherein each blob comprises a plurality of connected foreground blocks in the current frame;
(d) grouping the objects as one of: (i) new, (ii) persistent and (iii) recurring based on the number of foreground blocks and tracking count of the blobs;
(e) discarding the recurring objects.
2. The method of claim 1 , wherein distribution of the at least one feature data monitored for a period of time, is represented in a histogram.
3. The method of claim 2 , wherein a block is classified as background if the normalized distance of the block feature value from the mean of the histogram is smaller than a threshold or as foreground if larger.
4. The method of claim 1 , wherein a background map is generated for the current frame using the background blocks.
5. The method of claim 4 , where background of the scene is dynamically modeled using the background map over a plurality of frames.
6. The method of claim 5 , wherein the background model is used for inter-frame encoding of the sequence of video frames.
7. The method of claim 6 , wherein encoding bit-rate is lowered by skipping frames when no new object is detected in the scene.
8. The method of claim 1 , wherein an alarm is raised upon detection of salient objects.
9. An image processing apparatus comprising:
(a) a process configured to receive data from a sequence of video frames captured from a scene, wherein each frame comprises a plurality of pixels grouped in a plurality of image blocks;
(b) a processor configured to receive block feature data for each block, the block feature data being at least one of Sobel edge and spatial high frequency response values of each pixel averaged over the block; and
(c) for processor configured to detect salient objects in the scene by:
(i) classifying the blocks as background or foreground in a current frame based on temporal profile of the block feature data;
(ii) identifying objects as a plurality of blobs, wherein each blob comprises a plurality of connected foreground blocks in the current frame;
(iii) grouping the objects as one of: (i) new, (ii) persistent and (iii) recurring based on the number of foreground blocks and tracking count of the blobs;
(iv) discarding the recurring objects.
10. The apparatus of claim 9 , further comprising a background processor configured to generate a background map for the current frame using the background blocks.
11. The apparatus of claim 10 , wherein said background processor is configured to dynamically model background of the scene by using the background map over a plurality of frames.
12. The apparatus of claim 11 , wherein said background processor is configured to implement the background model in inter-frame encoding of the sequence of video frames.
13. A system for detecting salient objects in a scene, said system comprising:
(a) at least one stationary positioned capture device to capture a sequence of video frames from the scene, wherein each frame comprises a plurality of pixels grouped in a plurality of image blocks;
(b) an input interface configured for receiving pixel data and block feature data for each of the plurality of image blocks, the block feature data being at least one of Sobel edge and spatial high frequency response values of each pixel averaged over the block; and
(c) a processing unit configured for:
(i) classifying the blocks as background or foreground in a current frame based on temporal profile of the block feature data;
(ii) identifying objects as a plurality of blobs, wherein each blob comprises a plurality of connected foreground blocks in the current frame;
(iii) grouping the objects as one of: (i) new, (ii) persistent and (iii) recurring based on the number of foreground blocks and tracking count of the blobs;
(iv) discarding the recurring objects.
14. The system of claim 13 , wherein said processing unit is configured to generate a background map for the current frame using the background blocks.
15. The system of claim 14 , wherein said processing unit is configured to dynamically model background of the scene by using the background map over a plurality of frames.
16. The system of claim 15 , wherein said system further comprises an encoder to implement the background model in inter-frame encoding of the sequence of video frames.
17. The system of claim 16 , wherein the encoder applies SKIP mode compression by skipping frames where no new salient object is detected.
18. The system of claim 13 , wherein said system further comprises an alarm that is raised upon detection of salient objects.
19. The system of claim 18 , wherein said system further comprises recording means to store frames following raise of the alarm.
20. The system of claim 13 , wherein said system further comprises histograming means to monitor distribution of the at least one feature data for a period of time.