Robust RGB-T tracking via Bimodal Constrained Correlation Filtering

With the rapid development of Internet of Things, thermal camera has been widely applied in the intelligence surveillance. Thus, RGB-T tracker has become a hot topic, because it can effectively use the complement of RGB and thermal target information to achieve full-time visual tracking. To effectively enhance the RGB-T tracking speed, we propose a bimodal constrained correlation filter based method. The key point of our method is to integrate bimodal regularization into the traditional correlation filter tracking framework. In addition, to enhance the robustness of tracking performance, we mine the multi-source oriented correlation filter coefficients to make the regression parameters of RGB and thermal images avoid negative effect when obtaining complementary advantages. The experiment results demonstrate that the proposed method can improve the tracking speed without decreasing the tracking accuracy.