Multi-cue Integrated SVM Based Visual Tracking

Focusing on the accuracy requirement in visual tracking applications, this paper propose a multi-cue integration based classification method to track target in video sequence in the paper, which more suitable for multi-cue integration represent samples to classifier. Firstly the proposed tracking method, a pixel based initial samples construction method is designed to form a accurate classifier, a classifier oriented multi-cue integration method is proposed to represent samples to improve the robustness of tracking system against complex conditions. A weighted optimum method is proposed to locate the target in the output image for the purpose of realizing accurate tracking. And the robustness and accuracy of the proposed method are demonstrated by the experimental results and analysis on kinds of videos, the proposed tracking method resolved the limitation of training samples for the existing classification methods.