Distracter-aware tracking via correlation filter

Abstract

Visual object tracking is an attractive issue in the field of computer vision. Recently, correlation filters (CF) based trackers formulate the training process by solving the regression in the Fourier domain and show a great efficiency for tracking task. However, its efficiency collapsed when the distracters appear in the background. To improve the robustness of these trackers, we propose a distracter-aware tracking method via correlation filter, which utilizes the positions of target center and distracters. Since most of CF based trackers directly use the Gaussian-shaped label map as the regression target, which may lead the discriminability of tracker reduced. Differing from previous work, we first use the response map of CF model to detect the information of distracters and then utilize these information to design a distracter-aware label map as the regression target for the training process. To further promote the robustness of the tracker, we design a re-detection scheme which uses positive and negative filters to refine the tracking results when the distracters appear. The proposed method not only detect and capture the discriminative information to learn a distracter-aware label map for the filter training, but also utilize these information to train the positive and negative filters for object re-detection, which enhances the robustness of the tracker. We evaluate our proposed method on both the standard OTB2013 and OTB2015 benchmarks, the experimental results show the effectiveness and robustness of the proposed method.