Pedestrian traffic counting covers a variety of computer vision techniques such as pedestrian detection, pedestrian tracking, etc., while computer vision is also a research area of great interest in artificial intelligence. This technology can not only obtain the character information in the video, but also track the pedestrian movement and detect the number of pedestrians. As the current research field of pedestrian traffic detection is more limited to recognition and area detection, resulting in large errors and poor statistical results. Therefore, in this paper, we will improve the accuracy of pedestrian detection by calling yolo and Deep sort for multi-target detection through Open CV.