基于局部稀疏表示的目标跟踪

Object tracking based on local sparse representation

为了克服目标遮挡、姿态变换、光照变化等导致的视频目标跟踪失效问题,提出了一种自适应局部表观模型的跟踪方法.将目标分割成多个局部特征子块,选择其中部分具有相对显著性特征的子块,在粒子滤波框架下,使用稀疏表示方法独立跟踪选择的局部子块,通过各局部子块跟踪结果来估计目标.在跟踪过程中,为了保证跟踪子块的表观相对稳定,动态替换不稳定的局部特征块.实验结果表明:该方法与各种流行跟踪方法相比,跟踪结果稳定,特别是在目标部分遮挡和变形以及光照变化等条件下,具有更好的跟踪准确性.

To solve problems of object occlusion,posture and illumination changes during the process of object tracking,a novel tracking method based on adaptive local appearance model was presented.Object was segmented into many local feature sub blocks,and parts of them which are relative significant were chosen.Within the framework of particle filter,sparse representation was utilized to track every local sub block,and then object state was estimated by tracking results of sub blocks.During process of tracking,unstable local sub blocks were dynamically replaced to ensure the robustness of tracking.Experimental results demonstrate that our method outperforms existing tracking methods,and tracking results are more stable and precise especially under the condition of object part occlusion,shift and illumination changes.