Dynamic Saliency-Aware Regularization For Object Tracking

With a good balanced between tracking accuracy and speed,correlation filters?CF?has become one of the best object tracking frameworks,based on which many success-ful trackers have been developed.Recently,spatially regularized CF?SRDCF?has been developed to remedy the annoying boundary effects of CF tracking,thus further boost-s the tracking performance.However,SRDCF uses a fixed spatial regularization map constructed from a loose bounding box and its performance inevitably degrades when the target or background show significant variations,such as object deformation or oc-clusion.To address this problem,in this paper,we show that a static saliency-aware spatial regularization?SSAR?can help to further boost both the tracking accuracy and robustness.Next,we present to consider both frame saliency and spatial preference to online generate the temporal saliency-aware regularization?TSAR?for CF tracking.Fi-nally we propose a new dynamic saliency-aware regularization?DSAR?for CF tracking.With DSAR,a simple yet effective energy function,which reflects the object saliency and tracking reliability in spatial-temporal domain,is defined to guide the online up-dating of the DSAR map using an efficient level-set algorithm.Extensive experiments on standard benchmarks OTB-2013^[1],OTB-2015^[2]and VOT-2016^[3]validate that the proposed DSAR-CF leads to better performance in terms of accuracy and speed than CF tracking using the original spatial regularization,and outperform many existing state-of-the-art tracking methods.Finally,the ablation study,algorithm analysis,failure cases and speed analysis evaluate the effectiveness of the proposed method comprehensivly.