Time scale adaptive motion detection

1. A method for efficient motion detection of non-persistent objects comprising: selecting a video segment for evaluation;

evaluating said video segment to identify at least two first pixel classes wherein at least one of said first pixel classes corresponds to a plurality of stationary pixels and at least one of said first pixel classes corresponds to a plurality of pixels in apparent motion indicative of the presence of a non-persistent object in motion;

evaluating said video segment to identify at least two second pixel classes wherein at least one of said second pixel classes corresponds to a background and at least one of said second pixel classes corresponds to a foreground indicative of the presence of a non-persistent object;

wherein evaluating said video segment to identify at least two first pixel classes wherein at least one of said first pixel classes corresponds to a plurality of stationary pixels and at least one of said first pixel classes corresponds to a plurality of pixels in apparent motion indicative of the presence of an object in motion further comprises: identifying a plurality of candidate motion pixels corresponding to a plurality of pixels in apparent motion indicative of the presence of a non-persistent object in motion; and applying morphological filtering to said candidate motion pixels thereby providing a motion mask;

combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground, thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object; and providing an output indicating an object in motion.

2. The method of claim 1 wherein identifying a plurality of candidate motion pixels corresponding to a plurality of pixels in apparent motion indicative of the presence of an object in motion comprises at least one of: computing a frame-to-frame difference;

computing a double frame-to-frame difference; computing a multiple frame-to-frame difference;

computing an optical flow; and computing a block based motion estimation.

3. The method of claim 1 wherein applying morphological filtering to said candidate motion pixels further comprises: defining a refined set of motion pixels indicative of the presence of a non-persistent object in motion; and applying morphological hole filling on said refined set of motion pixels thereby providing a motion binary mask.

4. The method of claim 1 wherein evaluating said video segment to identify at least two second pixel classes wherein at least one of said second pixel classes corresponds to a background and at least one of said second pixel classes corresponds to a foreground indicative of the presence of a non-persistent object further comprises:

identifying a current frame of said selected video segment and identifying a current background frame of said selected video segment;

comparing said current background frame of said selected video segment and said identified current frame of said video segment to identify a foreground of said current frame; and

producing a foreground mask.

5. The method of claim 4 wherein identifying a current background frame of said selected video segment comprises one of:

estimating a current background frame of said video segment wherein estimating a current background frame comprises at least one of:

performing a pixel-wise running average on a plurality of incoming frames;

performing a pixel-wise weighted average on a plurality of incoming frames;

performing a pixel-wise median computation on a plurality of incoming frames; and

computing a set of pixel-wise statistical models corresponding to said current background frame.

6. The method of claim 4 wherein producing a foreground binary mask further comprises:

applying morphological filtering on said identified foreground of said current frame; and

comparing a size of said identified foreground of said current frame against a threshold size, wherein if said size of said identified foreground of said current frame is greater than said threshold size, said foreground of said current frame provides a foreground binary mask.

7. The method of claim 1 wherein said first pixel class corresponding to a plurality of pixels in apparent motion comprises one of a motion binary mask and a motion non-binary mask; and

wherein said second pixel class corresponding to a foreground indicative of the presence of a non-persistent object comprises one of a foreground binary mask and a foreground non-binary mask.

8. The method of claim 7 wherein combining said set of motion pixels indicative of the presence of an object in motion and said at least one of said pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises one of:

applying a binary, pixel-wise AND operation on said motion binary mask and said foreground binary mask; and

applying a binary, pixel-wise OR operation on said motion binary mask and said foreground binary mask.

9. The method of claim 7 wherein combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises:

applying a pixel-wise operation on said motion non-binary mask and said foreground non-binary mask comprising at least one of:

averaging a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a weighted average of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a product of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

conducting an order-statistical combination of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask; and

thresholding a resulting mask from said pixel-wise operation to yield said final motion mask.

10. The method of claim 1 wherein combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object, further comprises:

identifying at least one binary motion region from said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion;

identifying at least one binary foreground region from said at least one of said second pixel classes corresponding to a foreground indicative of the presence of a non-persistent object; and

defining said final motion mask based on at least one of:

an absolute size of an overlapping region of said identified binary motion region and said identified binary foreground region;

a fractional size of an overlapping region of said identified binary motion region and said identified binary foreground region; and

a shape of an overlapping region of said identified binary motion region and said identified binary foreground region.

11. The method of claim 7 wherein combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises:

identifying at least one non-binary motion region from said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion associated with said motion non-binary mask;

deriving a first continuous-valued mask indicative of a pixel-wise confidence metric for each of said identified non-binary motion regions based on an associated first continuous-valued value for said non-binary motion region;

identifying at least one non-binary foreground region from said at least one of said second pixel classes corresponding to a foreground associated with said foreground non-binary mask;

deriving a second continuous-valued mask indicative of a pixel-wise confidence metric for each of said identified non-binary foreground regions based on an associated second continuous-valued value for said non-binary foreground region; and

defining said final motion mask based on at least one of:

an absolute size of an overlapping region of said identified non-binary motion region and said identified non-binary foreground region;

a fractional size of an overlapping region of said identified non-binary motion region and said identified non-binary foreground region; and

a shape of an overlapping region of said identified non-binary motion region and said identified non-binary foreground region.

12. A method for efficient motion detection of non-persistent objects comprising:

selecting a video segment for evaluation;

evaluating said video segment to identify at least two first pixel classes wherein at least one of said first pixel classes corresponds to a plurality of stationary pixels and at least one of said first pixel classes corresponds to a plurality of pixels in apparent motion indicative of the presence of a non-persistent object in motion further comprising:

identifying a plurality of candidate motion pixels corresponding to a plurality of pixels in apparent motion indicative of the presence of a non-persistent object in motion; and

applying morphological filtering to said candidate motion pixels thereby providing a motion mask;

evaluating said video segment to identify at least two second pixel classes wherein at least one of said second pixel classes corresponds to a background and at least one of said second pixel classes corresponds to a foreground indicative of the presence of a non-persistent object further comprising:

identifying a current frame of said selected video segment and identifying a current background frame of said selected video segment;

comparing said current background frame of said selected video segment and said identified current frame of said video segment to identify a foreground of said current frame; and

producing a foreground mask;

combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground, thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object; and

providing an output indicating an object in motion.

13. The method of claim 12 wherein said first pixel class corresponding to a plurality of pixels in apparent motion comprises one of a motion binary mask and a motion non-binary mask; and

wherein said second pixel class corresponding to a foreground indicative of the presence of a non-persistent object comprises one of a foreground binary mask and a foreground non-binary mask.

14. The method of claim 13 wherein combining said set of motion pixels indicative of the presence of an object in motion and said at least one of said pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises one of:

applying a binary, pixel-wise AND operation on said motion binary mask and said foreground binary mask; and

applying a binary, pixel-wise OR operation on said motion binary mask and said foreground binary mask.

15. The method of claim 13 wherein combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises:

applying a pixel-wise operation on said motion non-binary mask and said foreground non-binary mask comprising at least one of:

averaging a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a weighted average of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a product of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

conducting an order-statistical combination of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask; and

thresholding a resulting mask from said pixel-wise operation to yield said final motion mask.

16. A system for efficient motion detection of non-persistent objects comprising: a camera configured to collect video data; a processor; and a computer-usable medium embodying computer code, said computer-usable medium being coupled to said processor, said computer code comprising non-transitory instruction media executable by said processor configured for:

selecting a video segment for evaluation;

evaluating said video segment to identify at least two first pixel classes wherein at least one of said first pixel classes corresponds to a plurality of stationary pixels and at least one of said first pixel classes corresponds to a plurality of pixels in apparent motion indicative of the presence of a non-persistent object in motion; evaluating said video segment to identify at least two second pixel classes wherein at least one of said second pixel classes corresponds to a background and at least one of said second pixel classes corresponds to a foreground indicative of the presence of a non-persistent object; wherein said first pixel class corresponding to a plurality of pixels in apparent motion comprises one of a motion binary mask and a motion non-binary mask; and wherein said second pixel class corresponding to a foreground indicative of the presence of a non-persistent object comprises one of a foreground binary mask and a foreground non-binary mask; by applying morphological filtering combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground, thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object; and

providing an output indicating an object in motion.

17. The system of claim 16 wherein combining said set of motion pixels indicative of the presence of an object in motion and said at least one of said pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises one of: applying a binary, pixel-wise AND operation on said motion binary mask and said foreground binary mask; and applying a binary, pixel-wise OR operation on said motion binary mask and said foreground binary mask.

18. The system of claim 16 wherein combining said at least one of said first pixel classes corresponding to a plurality of pixels in apparent motion and said at least one of said second pixel classes corresponding to a foreground of said selected video segment thereby defining a final motion mask in said selected video segment indicative of the presence of a non-persistent object further comprises:

applying a pixel-wise operation on said motion non-binary mask and said foreground non-binary mask comprising at least one of:

averaging a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a weighted average of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

computing a product of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask;

conducting an order-statistical combination of a continuous-value associated with said motion non-binary mask and a continuous-value associated with said foreground binary mask; and

thresholding a resulting mask from said pixel-wise operation to yield said final motion mask.