With the rapid growth of computer vision technology, the application of multi-object tracking (MOT) is becoming increasingly widespread. However, in practical applications, there are still unresolved issues with target occlusion. This article is based on YOLOv5 (You Only Look Once version 5) anti-occlusion multi-target tracking method, and designs a multi-target tracking model to improve the accuracy and robustness of multi-target tracking in complex environments. This article first provides an overview of the relevant research on multi-target tracking and deeply analyzes the impact of target occlusion on tracking performance. Subsequently, the basic principles of the YOLOv5 object detection algorithm are discussed in detail. Based on this, a multi-objective tracking framework is designed that combines YOLOv5 and re-detection mechanism. This framework fully utilizes the object detection capability of YOLOv5 to accurately identify and locate target objects in video frames. Its innovation lies in the application of a re-detection mechanism, which can timely re-detect and restore tracking even if the target is occluded. The framework also enhances multi-scale detection capabilities by adopting more effective feature pyramid networks. This network structure can better capture targets of different scales, enabling algorithms to detect small and large targets more accurately while maintaining high speed. By incorporating a self-adaptive anchor adjustment mechanism for different target shapes and sizes, the flexibility and adaptability of the algorithm are improved. Afterward, this article conducts testing and analysis on the model, and the test results showed that the model can maintain a tracking accuracy of over 75%, with accuracy data fluctuating around the benchmark value of 85%. This result fully demonstrates the superior performance of our model in anti-occlusion multi-target tracking.