Robust visual tracking via occlusion detection based on staple algorithm
Xiaoguang Niu, Xingqi Fang, Yu Qiao · 2017
In this paper, we propose a novel algorithm to handle occlusions for robust visual object tracking. For a robust and adaptive tracking algorithm, it is critical to distinguish occlusions from target appearance variations. Occlusions are introduced in the 3-D scene to 2-D plane projection process and usually correspond to interactions between the target and the background. In our algorithm, we exploit temporal and spatial context information to identify the occurrence of occlusions. We implement the proposed occlusion detection scheme based on Staple algorithm [2], which is an efficient and accurate tracking algorithm. Besides the tracker to estimate the state of the target, we also investigate the background patches around the target with trackers to monitor their interaction with the target. If occlusions are detected, the current target model stops updating. Comprehensive experiments on benchmark OTB-2013 show that our strategy can detect occlusions correctly and therefore enable the proposed tracking algorithm robust to occlusions.