An adaptive enhanced windowed correlation filter for visual tracking

Boundary effect, as an inherent drawback of discriminative correlation filter (DCF) trackers, cannot be handled well in most existing studies. This paper proposes an adaptive enhanced windowed correlation filter tracker (AEWCF), which can alleviate boundary effect adaptively and capture more reliable target information effectively. Firstly, a target enhanced likelihood map is introduced to extract reliable target information from searching window. Then, this paper further proposes an adaptive parted searching window including two sub-windows: the object region window depended on object likelihood maximization strategy and the background window generated by background suppression scheme. The extensive evaluation on OTB2015 benchmarks demonstrate that the AWCF tracker performs favorably against some popular trackers and keeps real-time speed.