After several years of development, the multi-target tracking algorithm has significantly transitioned from being researched to being put into practical production and life. The application field of human detection and tracking technology is closely related to our daily life. In order to solve the problems of the background complexity, the diversity of object shapes in the application of multi-target algorithms, and the mutual occlusion between multiple tracking targets and the lost target, this paper improves the DeepSORT target tracking algorithm, uses the improved YOLO network to detect pedestrians, inputs the detection frame to the Kalman filter for prediction output, and then uses the Hungarian algorithm to realize a tracking frame and detection frame of the predicted output. The experimental results show that target tracking accuracy is increased by 4.3%, the running time is the shortest, and the number of successfully tracked targets is relatively high.