Visual object tracking with scale and orientation adaptation

1. A method of tracking an object that appears in a plurality of image frames, comprising:

(a) dividing an identified object of one of the plurality of image frames into a plurality of object segments;

(b) tracking a location of each of the plurality of object segments in the image frame;

(c) estimating at least one of scale and orientation of the object using the location of each of the plurality of object segments; and

(d) obtaining position of the object using the estimated scale and orientation.

2. The method of claim 1 , comprising repeating steps (a)-(d) for each of the plurality of image frames.

3. The method of claim 1 , wherein the plurality of object segments comprise overlapping object segments.

4. The method of claim 1 ,

wherein each of the plurality of object segments is represented using a histogram; and

wherein the tracking includes tracking the location of each of the plurality of object segments using a mean shift vector of each of the plurality of object segments.

5. The method of claim 4 , wherein the tracking further includes:

characterizing visual features of each of the plurality of object segments through a statistical distribution in each of first and second image frames;

computing a similarity coefficient for each of the plurality of object segments based upon the statistical distribution in the first and second image frames; and

obtaining the location of each of the plurality of object segments in the second image frame using the corresponding similarity coefficient.

6. The method of claim 5 , wherein the visual features comprise a color and/or a texture of the object segment.

7. The method of claim 1 , wherein the tracking includes maintaining coherence between adjacent object segments using a Kalman filter, and

wherein each of a state transition matrix and a covariance matrix of the Kalman filter comprises a non diagonal matrix.

8. The method of claim 1 , wherein the tracking incldes:

identifying outlier object segments and in-lying object segments based upon the location of object segments in successive image frames and a pre-determined threshold; and

fitting a blob representing the object using the identified in-lying object segments to obtain the position of the object.

9. The method of claim 8 , wherein the identifying includes:

comparing similarity coefficient for each of the plurality of object segments with a pre-determined threshold value for identifying outlier object segments; and

re-initializing each of the plurality of object segments if a number of the plurality of outlier object segments having a similarity coefficient below the threshold value for indentifying outlier object segments is greater than the pre-determined threshold value therefore.

10. The method of claim 8 , wherein the fitting includes:

computing centralized image moments for the blob representing the object; and

computing a centroid and a covariance matrix using the centralized image moments; and

wherein the estimating includes determining the scale and orientation of the object using the centralized image moments and the covariance matrix.

11. A method of tracking an object that appears in a plurality of image frames, comprising:

identifying the object in one of the plurality of image frames;

dividing the identified object into a plurality of object segments;

estimating a location of each of the plurality of object segments using a mean shift vector of each of the plurality of object segments;

updating the estimated location of each of the plurality of object segments through a Kalman filter; and

estimating at least one of scale and orientation of the object using the updated location of each of the plurality of object segments.

12. The method of claim 11 ,

wherein each of the plurality of object segments is represented using a histogram; and

wherein the estimating includes tracking the location of each of the plurality of object segments in each of the plurality of image frames by minimizing a distance between histograms of successive image frames.

13. The method of claim 11 , wherein the estimating includes:

identifying in-lying object segments based upon an updated location of each of the plurality of object segments in successive frames and a pre-determined threshold; and

fitting a blob using the identified in-lying object segments to obtain the position of the object.

14. The method of claim 13 , wherein the fitting includes:

computing first and second image moments using a center of each of the identified in-lying object segments;

determining a major axis, a minor axis and an orientation angle using the first and second image moments; and

aligning each of the plurality of object segments along the orientation angle.

15. A system for tracking an object, comprising:

a memory configured to store a plurality of image frames with at least one object appearing in each of the plurality of image frames;

an image processing circuit configured to estimate a scale and an orientation of the at least one object in each of the plurality of image frames, wherein the image processing circuit comprises:

a plurality of tracking modules configured to track a position of each of a plurality of object segments of the at least one object; and

a computational module configured to estimate the scale and orientation of the at least one object using the position of each of the plurality of object segments.

16. The system of claim 15 , wherein the image processing circuit is configured to:

identify the at least one object in each of the plurality of image frames; and

divide the identified object into the plurality of object segments.

17. The system of claim 16 , wherein the plurality of object segments comprise overlapping object segments and wherein a size of each of the object segment is about 35% of the size of the at least one object.

18. The system of claim 15 , wherein each of the plurality of tracking modules comprises a mean shift based tracker, or an integral histogram based tracker, or combinations thereof.

19. The system of claim 15 , wherein the computational module comprises a Kalman filter configured to maintain coherence between adjacent object segments and to detect outlier object segments based upon location of object segments in successive frames and a pre-determined threshold.

20. The system of claim 19 , wherein diagonal elements of a state transition matrix of the Kalman filter are assigned a weight (w), wherein diagonal elements of a covariance matrix of the Kalman filter are assigned a weight (1−w)/(N−1) and wherein the weight (w) has a value between 0 and about 1.