Distractor-aware Visible and Infrared Tracking based on Multi-feature Fusion

It is still challenging for tracking the object of interest in complex environments. Combinations of different features can extract more discriminative information for the tracking problems. Fusion of the visible and infrared features is a feasible way since they are complementary. In this paper, a distractor-aware visible and infrared tracking method is presented. The multi-feature distractor-aware object model is designed by combining the information of the object, the distractors, the color feature of the visible image, and the intensity feature of the infrared image. Two probability map fusion methods are presented including the fixed weights and the adaptive weights based on the peak to sidelobe ratio (PSR). A scale updating strategy based on feature selection is also designed. Experiments on the public Object Tracking and Classification Beyond the Visible Spectrum (OTCBVS) datasets are conducted. The results demonstrate that the proposed tracking methods perform well in terms of accuracy and robustness in real surveillance scenes and show better results compared to several state-of-the-art methods.