Currently, detection-based multi-target tracking algorithms have shown excellent performance in dealing with the problems of missed detection and ID switching in complex scenes. However, these algorithms still need to be improved to face the performance limitations in the detection phase of moving targets in complex scenes. To address this challenge, a new convolutional hybrid attention mechanism is introduced in this paper. This mechanism strengthens the attention to sparse spatial deformation and contextual information in highly dynamic scenes through the hybrid attention model, and achieves the dynamic weighting of feature deformation at different scales. Further, a two-phase multi-target tracking method, CHAMTrack, is proposed in this paper. By using this attention mechanism in the detection phase of moving targets, the algorithm can enhance its ability to capture key information in complex multi-target tracking scenarios, and improve the tracking effect of targets of different scales in the same scene significantly. In order to verify the effectiveness of the proposed method, this paper tests it on MOT17 and MOT20 datasets respectively, and analyses of the results show that CHAMTrack significantly improves on the key performance indicators MOTA and IDSw. The ablation experiments further demonstrate its application value in the field of multi-target tracking in complex scenes.