Method and apparatus for detecting people within video frames based upon multiple colors within their clothing

1. A method, performed in a video analytic device, for detecting people within frames of video based upon multiple colors within their clothing, the method comprising:

receiving a current frame of video;

determining that a first color region within the current frame matches a first color of interest for a clothing uniform, wherein the determining is based on a first set of color representation constraints;

determining that a second color region within the current frame matches a second color of interest for the clothing uniform, wherein the determining is based on a second set of color representation constraints, wherein the first and second colors of interest are different; and

applying a set of geometric constraints to the first and second color regions to determine a count of people within the current frame wearing the clothing uniform,

wherein the first set of color representation constraints comprises a first color label and a first color classifier that represent the first color of interest, and the second set of color representation constraints comprises a second color label and a second color classifier that represent the second color of interest,

wherein determining that the first color region within the current frame matches the first color of interest and determining that the second color region within the current frame matches the second color of interest comprise:

determining, within a first resolution image of the current frame, a first set of pixels that match the first color label and a second set of pixels that match the second color label;

applying, within a second resolution image of the current frame, the first color classifier to the first set of pixels to classify the first set of pixels to determine the first color region, wherein the second resolution is higher than the first resolution; and

applying, within the second resolution image of the current frame, the second color classifier to the second set of pixels to classify the second set of pixels to determine the second color region.

2. The method of claim 1 , wherein the first and second color labels each comprise a color space that is selected from a plurality of color spaces based upon the particular color of interest represented by the color label.

3. The method of claim 2 , wherein the plurality of color spaces comprises a normalized Red, Green, and Blue color space and a Hue, Saturation, and Intensity color space.

4. The method of claim 2 , wherein the first and second color classifiers are determined using a color histogram.

5. The method of claim 1 further comprising determining a color detection region within the current frame, wherein the color detection region comprises a portion of the entire current frame and includes the first color region and the second color region, wherein the color detection region in the current frame corresponds to detected color regions within a preceding frame of the current frame.

6. The method of claim 1 further comprising analyzing a count of people for each of a plurality of frames of the video, including the current frame, to determine a final count of people for each of the plurality of frames.

7. The method of claim 6 , wherein the final count of people for each of the plurality of frames is a mode of the counts of people for the plurality of frames.

8. The method of claim 1 , wherein the first and second color labels each comprise a color space that is selected from a plurality of color spaces, wherein selecting the color space for each color label comprises:

annotating, within a training frame of video, pixels for each color of interest;

determining a compactness within a set of color spaces from the pixels for each color of interest; and

selecting the color space from the set of color spaces based on the compactness within the set of color spaces.

9. The method of claim 8 further comprising:

deriving a set of statistical measures from the pixels for each color of interest; and

training the color label for each color of interest, wherein the color label comprises the set of statistical measures and the selected color space.

10. The method of claim 9 , wherein the set of statistical measures comprises a color mean and a color distance.

11. The method of claim 8 further comprising:

extracting, in the selected color space, an invariant color feature for each of the pixels for each color of interest; and

training a classifier for each color of interest based on the extracted invariant color feature.

12. The method of claim 11 , wherein the invariant color feature is a color histogram and the classifier is a Support Vector Machine (SVM) classifier.

13. The method of claim 1 , wherein the set of geometric constraints comprises at least one of a minimum size constraint, a maximum size constraint, or a color region relationship.

14. A non-transient computer-readable storage element having computer readable code stored thereon for programming a computer to perform a method for detecting people within frames of video based upon multiple colors within their clothing, the method comprising:

receiving a current frame of video;

detecting, within a first resolution image of the current frame, multiple colors of interest to identify a first set of pixels of interest;

classifying, within a second resolution image of the current frame, colors of a second set of pixels of interest to identify pixels of the multiple colors of interest, wherein the second set of pixels of interest correspond to the first set of pixels of interest, and wherein the second resolution is higher than the first resolution;

determining a set of color regions within the second resolution image, wherein each of the color regions within the set contains at least one of the pixels of the multiple colors of interest; and

verifying that the set of color regions conform to a set of geometric constraints to determine a count of people, within the current frame, wearing a same type of clothing having the multiple colors of interest.

15. The storage element of claim 14 , wherein each color of interest is associated with a color label, wherein detecting the multiple colors of interest comprises determining that each of the pixels in the first set of pixels of interest matches one of the color labels, wherein each color label comprises a color space, a color mean, and a color distance.

16. The storage element of claim 14 , wherein classifying comprises applying a color classifier for each color of interest to each pixel in the second set of pixels of interest, wherein the color classifier is determined using an invariant color feature.

17. The storage element of claim 14 , wherein the method further comprising determining a color detection region within the current frame, wherein the color detection region comprises the first set of pixels of interest, the second set of pixels of interest, and the color regions, wherein the color detection region in the current frame corresponds to detected color regions within a preceding frame of the current frame.

18. The storage element of claim 14 , wherein the method further comprising:

determining a mode of a count of people over a plurality of frames of the video, including the current frame; and

selectively setting the mode as a final count of people for the plurality of frames.

19. The storage element of claim 14 , wherein the set of geometric constraints comprises at least one of a minimum size constraint, a maximum size constraint, or a color region relationship.