Color-Based Visual Object Tracking with Prediction and Error Judgment
Zhicheng Li, Bing Qiao, Shaobin Deng · 2009
Visual tracking algorithm plays an important role in the fields such as guidance and surveillance of mobile robot. These applications require the vision algorithm with strong robustness and fast processing speed. Camshift uses color histogram as a characteristic and mean shift as the search algorithm. The direction of grads ascension is used to reduce the characteristic match time, so that the object orientation could be faster. But, Camshift algorithm will fail when the background has the similar color with the target, it also cannot work correctly when occlusion occurs. This paper describes a new color based tracking algorithm in order to improve the theoretic limitation of Camshift. Firstly, the improved algorithm combines morphologic to solve the problems of divergence. Secondly, Kalman filter is used to predict the mass center point when the target has occlusion. Thirdly, it can receive error information from the error judgment analysis system to adjust the tracking. Finally, the result of this experiments shows that the proposed algorithm can track fast moving object successfully and have better robustness for occlusion.