An Automatic Object Attention Likelihood Map Correlation Filter for Visual Tracking

Discriminative correlation filter-based trackers have achieved excellent performance. However, they are still suffered from the inherent boundary effect. To alleviate it, this paper proposes an automatic object attention likelihood map correlation filter. Specifically, an automatic object attention searching window is developed to highlight the object more precisely and suppress the background more appropriately, which divides the searching area into three regions and applies different object attention strategies on each area. Furthermore, combining it with the spatially regularized discriminative correlation filter, we propose the automatic object attention likelihood map correlation filter tracker. It further alleviates the boundary effect and adaptively generates a more precise tracking model frame by frame. The experimental evaluations on the popular OT2015 benchmark demonstrate that our proposed tracker gains outstanding tracking performance and offers a satisfying robustness to complex tracking scenarios.