Improved CAMShift object tracking based on Epanechnikov Kernel Density Estimation and Kalman filter
Dawei Li, Lihong Xu, Yang Wu · 2017
This paper first chooses Epanechnikov Kernel Density Estimation for moving foreground detection. Targets are extracted by labeling connected regions in the detected binary image. Kalman filter is then employed to proffer initial searching window for CAMShift algorithm to track targets and predict the future position of the targets. Meanwhile, the histogram of target is updated by the color information in the region obtained by periodical detection of Epanechnikov Kernel Density Estimation. This paper also addresses the occlusion problem in tracking. Experimental results show that the strategy which combines the KDE foreground detection, Kalman filter and CAMShift, can realize automatic and efficient tracking of moving target. The reliable performance of this algorithm satisfies the real-time requirement, and is robust against the effects of unstable scene illumination, and object occlusion.