Discriminative Visual Tracking Using Multifeature and Adaptive Dictionary Learning
We propose a novel tracking method using color features and texture features to obtain accurate target appearance model.Each feature dictionary is independent and learned by labeled consistent K-SVD.In the subsequent frames,we exploit the maximum similarity of sparse features by the minimal reconstruction error criterion to locate the best tracking result.When a significant change occurs,we propose an adaptive dictionary learning which update the background template incrementally using the positive and negative samples of the current target.We compared our method with the existing techniques in OTB100 and VOT2017 dataset,and the experimental results show that our proposed method achieved substantially better performance.
Visual tracking Sparse representation Multiple features Adaptive updating
Penggen Zheng Jin Zhan Huimin Zhao Jujian Lv
School of Computer Science,Guangdong Polytechnic Normal University,Guangzhou,China;Guangzhou Key Laboratory of Digital Content Processing and Security Technologies,Guangzhou,China
国际会议
广州
英文
221-232
2018-11-23（万方平台首次上网日期，不代表论文的发表时间）