Multiscale Dynamic Perception Network for RGBT Tracking

Aiming at the problems of inefficient modal feature fusion and sensitive to target scale change in visible light-thermal infrared (RGBT) target tracking, an RGBT target tracking method based on multi-scale dynamic perception network (MDPNet) is proposed. Taking MDNet as the baseline network, a dual-stream feature extraction branch is constructed to extract the features of visible and thermal infrared modalities respectively. Firstly, the local details and global context information of the target are captured by the multi-scale synergy module; secondly, the dynamic perception fusion module is introduced in the feature fusion stage, which realizes the dynamic selection and fusion of modal features through the adaptive channel weighting mechanism. Experiments on the RGBT234 dataset and LasHeR dataset show that the proposed method has significant advantages in complex scenarios such as occlusion and scale change, which verifies the effectiveness of the multi-scale dynamic perception and cross-modal interaction mechanism.