A background-aware correlation filter with adaptive saliency-aware regularization for visual tracking

Recently, the discriminative correlation filters (DCF)-based methods have performed excellent precision and speed in object tracking. Due to the continuous change and expansion of the search region, the problem of insufficient training samples is solved by the periodicity hypothesis, which inevitably introduces boundary effects that can lead to severe failures in the detection stage. In this paper, we firstly add a background penalty factor into the correlation filter and propose a novel spatial regularization term by using the saliency detection method. Based on the above two points, a background-aware correlation filter model with saliency-aware regularization is established. Secondly, in order to solve the model better and faster, we introduce an energy function for the solution of the spatial weight and apply the alternating direction method of multipliers (ADMM) method and deduce the closed-form solution of each subproblem of the objective function efficiently. Thirdly, we propose an adaptive updating mechanism based on the variation of target appearance and the reliability of tracking results, which can update the model online by adjusting the spatial weight distribution for precisely tracking in the spatio-temporal domain. Finally, we apply two BAASR models to estimate the position and the scale of the target, respectively. One model adopts hand-crafted features at multiple scales to select the optimal scale, while the other model predicts the optimal position by fusing hand-crafted features with deep features extracted from the trained network models. Extensive experiments are carried out on the following five datasets, OTB-2013, OTB-2015, UAV123, UAV20L, and TC128. Experimental results demonstrate that our tracker has superior robustness and performance.

Automated summary

In this paper, a DCF-based object tracking method is proposed, which introduces background and saliency-aware spatial regularization to solve the problem of insufficient training samples and alleviate boundary effects. The ADMM method is used to efficiently solve the objective function. An adaptive updating mechanism is also introduced for online model updating. Experimental results demonstrate that the proposed method achieves superior robustness and performance.