An Adaptive Implementation of the Kernel-Based Object Tracking Method

Object tracking is one of the critical tasks in computer vision. The kernel-based object tracking (KBOT), employed an isotropic kernel to spatially mask the feature histogram-based target representations, is attractive for its ability toward to a real-time object tracking. In this paper, an adaptive dynamic updating principle of target model is proposed to improve the algorithm. Experiment results in implementation of the improved algorithm shown that the new method can improve the performance of the KBOT. Not only can it successfully cope with camera motion, background clutter, and target partial occlusions, rotation, scale variations, but also can be applied to rigid objects as well as nonrigid objects in visual tracking