Improved adaptive template updating strategy based on correlation filter in tracking

Abstract

Linear interpolation is adopted to update model with a fixed learning rate in target tracking. The traditional template update method is not satisfactory when dealing with complex environments. In order to prevent losing the target and improve the robustness, this paper creatively uses the NPSR (normalized peak side lobe ratio) to establish a target occlusion judgment mechanism. Taking the NPSR as the confidence, the weights of all historical templates are set according to the confidence. Therefore, the filtering template with the highest local historical reliability is fused with the original update mechanism. Then, the learning rate in the template update process is adaptively adjusted according to the current state of the target. Based on the OTB100 datasets, the improved adaptive template update strategy is applied to the KCF (Kernel Correlation Filter) tracking algorithm. The results show that our method has important research and application value for the correlation filter tracking algorithm.