Kernel-Bandwidth Adaptation for Tracking Object Changing in Size
In the case of tracking object changing in size, traditional mean-shift based algorithm always leads to poor localization owing to its unchanged kernel-bandwidth. To overcome this limitation, a novel kernel-bandwidth adaptation method is proposed where object affine model is employed to describe scaling problem. With the registration of object centroid in consecutive frames by backward tracking, scaling magnitude in the affine model can be estimated with more accuracy. Therefore, kernel-bandwidth is updated with respect to the scaling magnitude so as to keep up with variety of object size. We have applied the proposed method to track vehicles changing in size with encouraging results.
KeywordsConsecutive Frame Affine Model Poor Localization Scaling Magnitude Bhattacharyya Coefficient
Unable to display preview. Download preview PDF.
- 4.Bradski, G.R.: Computer vision face tracking for use in a perceptual user interface. In: IEEE Workshop on Applications of Computer Vision, Princeton, NJ pp. 214–219.(1998)Google Scholar
- 6.Collins, R.T.: Mean shift blob tracking through scale space. IEEE Int. Proc. Computer Vision and Pattern Recognition 2, 234–240 (2003)Google Scholar
- 7.Hu, W., Wang, S., Lin, R.S., Levinson, S.: Tracking of object with SVM regression. IEEE Int. Conf. Computer Vision and Pattern Recognition 2, 240–245 (2001)Google Scholar
- 8.Olson, C.F.: Maximum-Likelihood template matching. IEEE Int. Conf. Computer Vision and Pattern Recognition 2, 52–57 (2000)Google Scholar