Vision based target tracking that distinguishes facial feature targets

1. A facial recognition method comprising:

initializing target position and scale within a video frame;

extracting positive and negative face pair samples from candidate face targets in the video frame;

extracting higher-dimensional multi-pixel Haar-like facial features, including pixel intensities, from the positive and negative face pair samples;

using a sparse coding function reducing dimension of the higher-dimensional multi-pixel Haar-like facial features, including:

selecting a sub-set of more discriminative Haar-like facial features from among the higher-dimensional multi-pixel Haar-like facial features; and

forming a sparse feature matrix from the sub-set of more discriminative Haar-like facial features, including:

forming columns linking non-zero items in a solution vector to the sub-set of more discriminative Haar-like facial features; and

deleting columns where the corresponding item in the solution vector is zero; and

classifying targets based on content of the sparse feature matrix.

2. The facial recognition method of claim 1 , wherein classifying targets comprises classifying targets using a naive Bayesian classifier.

3. The facial recognition method of claim 1 , wherein classifying targets comprises:

classifying and scoring results derived from the sub-set of more discriminative Haar-like features to locate a target, from among candidate face targets, in the video frame; and

determining similarity of the target to a target template.

4. The facial recognition method of claim 3 , wherein determining similarity of the target to a target template comprising determining similarity of the target to the target template using a naive Bayesian classifier.

5. The facial recognition method of claim 3 , further comprising:

determining that the target is dissimilar to the target template; and

replacing the target template with the target based on the determined dissimilarity.

6. The facial recognition method of claim 3 , further comprising:

determining that the target is similar to the target template; and

updating a Bayesian classifier based on the determined similarity.

7. The facial recognition method of claim 3 , further comprising accessing the video frame; and

wherein extracting higher-dimensional multi-pixel Haar-like features comprises extracting higher-dimensional multi-pixel Haar-like features from candidate face targets in the video frame.

8. The facial recognition method of claim 3 , wherein classifying and scoring results derived from the sub-subset or more discriminative Haar-like features comprises classifying the results using a binary classification process.

9. The facial recognition method of claim 1 , further comprising accessing the video frame; and

wherein classifying targets comprises locating a target in the video frame based on content in the sparse feature matrix.

10. The facial recognition method of claim 1 , wherein extracting positive and negative face pair samples from the candidate face targets in the video frame comprises extracting the positive and negative samples in a neighborhood around candidate face target positions within the video frame.

11. The facial recognition method of claim 1 , wherein extracting higher-dimensional multi-pixel Haar-like features comprises:

detecting an eye region and a cheek region of a candidate face target at a target position; and

determining a multi-pixel Haar-like feature that includes adjacent triangles above the eye region and the check region within the video frame.

12. A computer system comprising:

a processor;

system memory coupled to the processor and storing instructions configured to cause the processor to:

initialize target position and scale within a video frame;

extract positive and negative face pair samples from candidate face targets in the video frame;

extract higher-dimensional multi-pixel Haar-like facial features, including pixel intensities, from the positive and negative face pair samples;

use a sparse coding function reducing dimension of the higher-dimensional multi-pixel Haar-like facial features, including:

selecting a sub-set of more discriminative Haar-like facial features from among the higher-dimensional multi-pixel Haar-like features; and

form a sparse feature matrix from the sub-set of more discriminative Haar-like facial features, including:

form columns linking non-zero items in a solution vector to the sub-set of more discriminative Haar-like facial features; and

delete columns where the corresponding item in the solution vector is zero; and

classify targets based on content of the sparse feature matrix.

13. The computer system of claim 12 , further comprising instructions configured to access the video frame; and

wherein instructions configured to classify targets comprise instructions configured to locate a target in the video frame based on content in the sparse feature matrix.

14. The computer system of claim 12 , wherein instructions configured to extract positive and negative face pair samples from candidate face targets in the video frame comprise instructions configured to extract the positive and negative face pair samples in a neighborhood around candidate face target positions within the video frame.

15. The computer system of claim 12 , wherein instructions configured to extract higher-dimensional multi-pixel Haar-like features comprise instructions configured to:

detect an eye region and a cheek region of a candidate face target at a target position; and

determine a multi-pixel Haar-like feature that includes adjacent triangles above the eye region and the check region within the video frame.

16. The computer system of claim 12 , wherein instructions configured to classify targets comprise instructions configured to:

classify and score results derived from the sub-set of more discriminative Haar-like features to locate a target, from among candidate face targets, in the video frame; and

determine similarity of the target to a target template.

17. The computer system of claim 16 , further comprising instructions configured to:

determine that the target is dissimilar to the target template; and

replace the target template with the target based on the determined dissimilarity.

18. The computer system of claim 16 , further comprising instructions configured to:

determine that the target is similar to the target template; and

updating a Bayesian classifier based on the determined similarity.

19. The computer system of claim 16 , wherein instructions configured to classify and score results derived from the sub-set of more discriminative Haar-like features comprise instructions configured to classify the results using a binary classification process.