Abstract: To enhance the tracking performance of vehicles and human targets under conditions of occlusion, background noise, and deformation, a multi-target tracking method based on the integration of static,dynamic target detection and feature point matching is proposed. Firstly, the Faster-CNN static detection region and a TFCD dynamic region are obtained separately, and joint detection is implemented to obtain the target of interest (TOI) region. Secondly, ORB feature points are extracted from each target region in current and its next frames, and then feature point matching is performed through a " neighbor window " optimization method to further achieve target matching. Tests on several publicly available experimental datasets show that the proposed method outperforms comparative methods in metrics such as IDF1 and SR, validating its effectiveness in multi-target tracking tasks, particularly in application scenarios prioritizing identity recognition.