Real-time Outdoor Abandoned Object Detection Algorithm based on Detection and Tracking
Ziteng Wang, Yeqin Shao · 2023
Aiming at the problem that the existing abandoned object detection algorithms cannot solve the problems of light and shadow changes and target occlusion, a real-time detection method of outdoor abandoned objects based on improved YOLOv5s and StrongSORT was proposed. Combining the advantages of SPPF, such as small computation and fast speed, with the rich information of C2f gradient flow, a novel spatial pyramid pooling structure SPPFC2FC is proposed. The trajectory of the detected object was tracked based on the target tracking algorithm, and the Euclidean distance was used to match the relationship between the object and the corresponding passenger, and the abandoned object was judged. The enhanced YOLOv5s exhibited a 1.1% increase in mAP0.5 and a 2.2% increase in mAP0.95. The experimental results show that the proposed method can accurately determine the outdoor abandoned objects even in complex scenes.