Research on pedestrian tracking detection based on YOLOv5 and improved DeepSORT

With the development and integration of big data and artificial intelligence, intelligent transportation, video surveillance, and other fields are receiving increasing attention. In the field of video surveillance, tracking targets is a challenging issue, especially in crowded environments where occlusion and false detection are prone to occur, and multi target detection frames are also prone to confusion.This article uses the YOLOv5 algorithm as the foundation for pedestrian tracking and detection, proposes an improved metric called DeepSORT, and extracts the skeleton network part in a lightweight manner. The network is further pruned, greatly reducing the number of parameters without reducing too much performance. The experimental results demonstrate that the proposed method improves the target recognition and tracking capabilities of pedestrian tracking and detection.