Studies on Vehicle Object Detection and Tracking in UAV Aerial Data

Studies on multi-object detection and tracking for vehicles are important research topics and have been widely used in various fields such as autonomous driving, anomaly detection, traffic monitoring, and intelligent transportation system. Because of the interference factors such as long-time occlusion and heavy traffic flow, there are two major limitations to accurate vehicle tracking: missed detections and false detections. In this paper, for the demand of accuracy and real-time vehicle object detection and tracking, the YOLO v5 and Deepsort are combined to realize object detection and tracking in unmanned aerial vehicle (UAV) data. The object detector is trained and validated on its own dataset of traffic collected from highways; the open-source vehicle deep feature training dataset is used to train for tracker weights. Finally, the experiments verify the vehicle multi-object detection and tracking function in the case of heavy traffic flow with different vehicle types. The proposed method in this paper has certain theoretical significance and practical application value in the field related to multi-object detection and tracking of moving vehicles.