MULTI-HIERARCHICAL INDEPENDENT CORRELATION FILTERS FOR VISUAL TRACKING

For visual object tracking, most of the traditional correlation filters (CF) based methods suffer from the bottleneck of feature redundancy and lack of motion information. In this paper, we design a novel tracking framework, called multi-hierarchical independent correlation filters (MHIT). The framework consists of hierarchical features selection, independent group CF online learning, adaptive multi-branch CF fusion and motion estimation module. Specifically, the multi-hierarchical deep features of CNN representing different semantic information can be fully employed to track multi-scale objects. To fully learn redundant deep features, each hierarchical feature is independently fed into a single branch to implement the online learning of parameters. Finally, an adaptive weight scheme is integrated into the framework to fuse these independent multi-branch CFs for robust visual object tracking. Furthermore, the motion estimation module is introduced to capture motion information, which effectively alleviates the problem of fast motion. Extensive experiments on OTB and VOT datasets show that the proposed MHIT tracker can significantly improve the tracking performance.