Adaptive Kernel Based Tracking Using Mean-Shift
The mean shift algorithm is an kernel based way for efficient object tracking. However, there is presently no clean mechanism for selecting kernel bandwidth when the object size is changing. We present an adaptive kernel bandwidth selection method for rigid object tracking. The kernel bandwidth is updated by using the object affine model that is estimated by using object corner correspondences between two consecutive frames. The centroid of object is registered by a special backward tracking method. M-estimate method is used to reject mismatched pairs (outliers) so as to get better regression results. We have applied the proposed method to track vehicles changing in size with encouraging results.
KeywordsObject Tracking Consecutive Frame Shift Algorithm Affine Model Kernel Bandwidth
Unable to display preview. Download preview PDF.
- 1.Comaniciu, D., Ramesh, V., Meer, P.: Real-time tracking of non-rigid objects using mean shift. IEEE Int. Proc. Computer Vision and Pattern Recognition 2, 142–149 (2000)Google Scholar
- 6.Harris, C., Stephens, M.: A combined corner and edge detector. In: Int. Conf. Alvey Vision Conference, pp. 189–192 (1988)Google Scholar
- 8.Collins, R.T.: Mean-shift blob tracking through scale space. IEEE Int. Proc. Computer Vision and Pattern Recognition 2, 234–240 (2003)Google Scholar
- 9.Hu, W., Wang, S., Lin, R.-S., Levinson, S.: Tracking of object with SVM regression. In: IEEE Int. Conf. Computer Vision and Pattern Recognition, vol. 2, pp. 240–245 (2001)Google Scholar
- 10.Olson, C.F.: Maximum-Likelihood template matching. In: IEEE Int. Conf. Computer Vision and Pattern Recognition, vol. 2, pp. 52–57 (2000)Google Scholar