Object Tracking Based on Camshift with Multi-feature Fusion
Zhiyu Zhou, Dichong Wu, Xiaolong Peng, Zefei Zhu, Kaikai Luo · Journal of Software · 2014
It is very hard for traditional Camshift to survive of drastic interferences and occlusions of similar objects. This paper puts forward an innovative tracking method using Camshift with multi-feature fusion. Firstly, SIFT features and edge features of the Camshift in RGB space are counted to reduce the probability of disruption by occlusion and clutter. Then, the texture features are collected to resolve the problems of analogue interference, the texture similarity between current frame and previous frames are calculated to determine the object area. The paper also describes the GM(1,1) prediction model, which could solve the occlusion problems in a novel way. Finally, through the motion trajectory, it can anticipate the exact position of the object. The results of several tracking tasks prove that our method has solved problems of occlusions, interferences and shadows. And it performs well in both tracking robustness and computational efficiency.