Background understanding in video data

1. A computer-implemented method for long-term understanding of background modeling, the method comprising executing on a processor the steps of:

determining dimensional gradient model values for pixel image data extracted from features of input video data;

defining average image pixel gradient model values for each pixel of the extracted pixel image data by averaging the determined dimensional gradient model values with the determined dimensional gradient model values of neighboring pixels as a function of their distance to the pixel; and

estimating a background static image scene of the input video data to comprise pixels of the extracted pixel image data that are not background pixel model pixels, and that have average image pixel gradient model values that are not less than a threshold distance value from the average image pixel gradient model values of the background model pixels, or that have average image pixel gradient model values that are less than the threshold distance value from the average image pixel gradient model values of the background model pixels but are closer to the average image pixel gradient model values of another foreground model set of the pixels.

2. The method of claim 1 , wherein the step of defining the average image pixel gradient model values comprises averaging the determined dimensional gradient model values with previous gradient values for each of the pixels.

3. The method of claim 1 , further comprising:

convolving the features pixel image data extracted from input video data with a horizontal filter to generate a horizontal dimension pixel gradient;

convolving the features pixel image data extracted from input video data with a vertical filter to generate a vertical dimension pixel gradient; and

defining the average image pixel gradient model values as a function of:

averages of the horizontal dimension pixel gradients with horizontal dimension pixel gradients of previous determined gradients of the respective image pixels, and the horizontal dimension pixel gradients of each of a plurality of neighboring pixels as a function of their respective distances of the neighboring pixels to the image pixel; and

averages of the vertical dimension pixel gradients with vertical dimension pixel gradients of any previous determined gradients of the respective image pixels, and with vertical dimension pixel gradients of the each of a plurality of neighboring pixels as a function of their respective distances of the neighboring pixels to the image pixel.

4. The method of claim 1 , further comprising:

integrating computer-readable program code into a computer system comprising a processor, a computer readable memory in circuit communication with the processor, and a computer readable hardware storage device in circuit communication with the processor; and

wherein the processor executes program code instructions stored on the computer-readable storage medium via the computer readable memory and thereby performs the steps of determining the dimensional gradient model values for pixel image data extracted from features of input video data, defining the average image pixel gradient model values for each pixel of the extracted pixel image data, and estimating the background static image scene of the input video data.

5. The method of claim 1 , further comprising:

defining a background pixel model to comprise a plurality of pixels of the pixel image data that have average image pixel gradient model values of a similar orientation and magnitude.

6. The method of claim 5 , further comprising:

determining if there is an existing background pixel model for ones of the pixels;

in response to determining that there is not an existing background pixel model for the ones of the pixels, initiating a first background pixel model for the ones of the pixels comprising their respective average image pixel gradient model values as a first model mean, and with an initial weight of one; and

in response to determining that there is an existing background pixel model, determining distances between the average image pixel gradient model values and the means of the existing model to identify a background pixel model of image pixel gradient model values having a smallest model distance, updating the mean of the identified background pixel model to an absolute average mean as a function of the image pixel averaged gradient mean value, and incrementing a weight of the identified background pixel model in response to a smallest model distance meeting a matching threshold requirement.

7. The method of claim 1 , further comprising:

updating weights of neighboring image pixels of the background model pixels with a factor defined by a decaying kernel function as a function of their distance to the image pixel; and

updating the background models to define the neighboring image pixels with largest updated weights as representing background pixel data, and remaining others of the neighboring image pixels that do not have the largest updated weights as representing foreground data.

8. The method of claim 7 , further comprising:

converting color image data of the extracted feature pixel image data to grey-scale image brightness data in response to the input video data color image data comprising more than one channel of color information.

9. A system, comprising:

a processing unit;

a computer readable memory in communication with the processing unit; and

a computer-readable hardware storage device in communication with the processing unit;

wherein the processing unit, when executing program instructions stored on the computer-readable storage medium via the computer readable memory:

determines dimensional gradient model values for pixel image data extracted from features of input video data;

defines average image pixel gradient model values for each pixel of the extracted pixel image data by averaging the determined dimensional gradient model values with the determined dimensional gradient model values of neighboring pixels as a function of their distance to the pixel; and

estimates a background static image scene of the input video data to comprise pixels of the extracted pixel image data that are not background pixel model pixels, and that have average image pixel gradient model values that are not less than a threshold distance value from the average image pixel gradient model values of the background model pixels, or that have average image pixel gradient model values that are less than the threshold distance value from the average image pixel gradient model values of the background model pixels but are closer to the average image pixel gradient model values of another foreground model set of the pixels.

10. The system of claim 9 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory, defines the average image pixel gradient model values as a function of averaging the determined dimensional gradient model values with previous gradient values for each of the pixels.

11. The system of claim 9 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory:

convolves the features pixel image data extracted from input video data with a horizontal filter to generate a horizontal dimension pixel gradient;

convolves the features pixel image data extracted from input video data with a vertical filter to generate a vertical dimension pixel gradient; and

defines the average image pixel gradient model values as a function of:

averages of the horizontal dimension pixel gradients with horizontal dimension pixel gradients of previous determined gradients of the respective image pixels, and the horizontal dimension pixel gradients of each of a plurality of neighboring pixels as a function of their respective distances of the neighboring pixels to the image pixel; and

averages of the vertical dimension pixel gradients with vertical dimension pixel gradients of any previous determined gradients of the respective image pixels, and with vertical dimension pixel gradients of the each of a plurality of neighboring pixels as a function of their respective distances of the neighboring pixels to the image pixel.

12. The system of claim 9 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory:

updates weights of neighboring image pixels of the background model pixels with a factor defined by a decaying kernel function as a function of their distance to the image pixel; and

updates the background models to define the neighboring image pixels with largest updated weights as representing background pixel data, and remaining others of the neighboring image pixels that do not have the largest updated weights as representing foreground data.

13. The system of claim 9 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory:

converts color image data of the extracted feature pixel image data to grey-scale image brightness data in response to the input video data color image data comprising more than one channel of color information.

14. The system of claim 9 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory defines a background pixel model to comprise a plurality of pixels of the pixel image data that have average image pixel gradient model values of a similar orientation and magnitude.

15. The system of claim 14 , wherein the processing unit, when executing the program instructions stored on the computer-readable storage medium via the computer readable memory:

determines if there is an existing background pixel model for ones of the pixels;

in response to determining that there is not an existing background pixel model for the ones of the pixels, initiates a first background pixel model for the ones of the pixels comprising their respective average image pixel gradient model values as a first model mean, and with an initial weight of one; and

in response to determining that there is an existing background pixel model, determines distances between the average image pixel gradient model values and the means of the existing model to identify a background pixel model having a smallest model distance, updates the mean of the identified background pixel model to an absolute average mean as a function of the image pixel averaged gradient mean value, and increments a weight of the identified background pixel model in response to a smallest model distance meeting a matching threshold requirement.

16. An article of manufacture for long-term understanding of background modeling, comprising:

a computer readable tangible hardware storage device having computer readable program code embodied therewith, the computer readable program code comprising instructions that, when executed by a computer processor, cause the computer processor to:

determine dimensional gradient model values for pixel image data extracted from features of input video data;

define average image pixel gradient model values for each pixel of the extracted pixel image data by averaging the determined dimensional gradient model values with the determined dimensional gradient model values of neighboring pixels as a function of their distance to the pixel;

define a background pixel model to comprise a plurality of pixels of the pixel image data that have average image pixel gradient model values of a similar orientation and magnitude; and

estimate a background static image scene of the input video data to comprise pixels of the extracted pixel image data that are not background pixel model pixels, and that have average image pixel gradient model values that are not less than a threshold distance value from the average image pixel gradient model values of the background model pixels, or that have average image pixel gradient model values that are less than the threshold distance value from the average image pixel gradient model values of the background model pixels but are closer to the average image pixel gradient model values of another foreground model set of the pixels.

17. The article of manufacture of claim 16 , wherein the computer readable program code instructions, when executed by the computer processor, further cause the computer processor to define the average image pixel gradient model values as a function of averaging the determined dimensional gradient model values with previous gradient values for each of the pixels.

18. The article of manufacture of claim 16 , wherein the computer readable program code instructions, when executed by the computer processor, further cause the computer processor to:

update weights of neighboring image pixels of the background model pixels with a factor defined by a decaying kernel function as a function of their distance to the image pixel; and

update the background models to define the neighboring image pixels with largest updated weights as representing background pixel data, and remaining others of the neighboring image pixels that do not have the largest updated weights as representing foreground data.

19. The article of manufacture of claim 16 , wherein the computer readable program code instructions, when executed by the computer processor, further cause the computer processor to define a background pixel model to comprise a plurality of pixels of the pixel image data that have average image pixel gradient model values of a similar orientation and magnitude.

20. The article of manufacture of claim 19 , wherein the computer readable program code instructions, when executed by the computer processor, further cause the computer processor to:

determine if there is an existing background pixel model for ones of the pixels;

in response to determining that there is not an existing background pixel model for the ones of the pixels, initiate a first background pixel model for the ones of the pixels comprising their respective average image pixel gradient model values as a first model mean, and with an initial weight of one; and

in response to determining that there is an existing background pixel model, determine distances between the average image pixel gradient model values and the means of the existing model to identify a background pixel model having a smallest model distance, update the mean of the identified background pixel model to an absolute average mean as a function of the image pixel averaged gradient mean value, and increment a weight of the identified background pixel model in response to a smallest model distance meeting a matching threshold requirement.