RGB-T object tracking via sparse response-consistency discriminative correlation filters

RGB-T object tracking is an emerging topic in the object tracking community owing to the complementarity of visual and thermal infrared images. However, existing RGB-T trackers usually focus on feature-level fusion and struggle to satisfy real-time requirements. In this study, we propose a real-time RGB-T object tracking algorithm based on sparse response-consistency discriminative correlation filters. First, to leverage the collaboration and heterogeneity of multi-modal images, we jointly learn discriminative correlation filters of visual and thermal infrared modalities via sparse response consistency. Second, we introduce an adaptive spatial regularization strategy to mitigate the boundary effects of the correlation filter-based tracking algorithm. Finally, to efficiently utilize the complementarity of visual and thermal infrared modalities, we design an adaptive decision-fusionand-updating strategy, which can reduce the adverse impact of unreliable modalities on tracking performance and boost the running speed. Extensive experiments on four large-scale RGB-T tracking benchmark datasets demonstrate that the proposed method performs favorably against state-of-the-art trackers in terms of precision and robustness, achieving a tracking speed of up to 53.447 fps.