A YOLO-X Deep Learning Approach to Detect Traffic Targets from UAV Video with On-Board Vehicle Data Validation
Junli Liu, Xiaofeng Liu, Qiang Chen, Dongpeng Yue · CICTP 2022 · 2022
The unmanned aerial vehicle (UAV) monitoring platform has the advantages of flexibility, traceability, convenient storage and wide-view, but the challenge is to extract the traffic parameters from the UAV-videos accurately. This paper proposes a YOLO-X deep learning approach to detect vehicles using the DJI Phantom 4RTK UAV and vehicle on-board unit. The UAV is used to monitor the traffic situation at various altitudes, and the vehicle on-board unit is used to record the accurate speed of the experiment vehicle. The YOLO-X deep learning framework and the video frame-based detection algorithm are incorporated and a field experiment was conducted. The model was trained based on collected and aerial data processed by the deep learning neural network, and compared with the ground vehicle data. Results show the accuracy of target detection is 92.58%, and the target vehicle speed detection is 94%. This demonstrates the proposed approach is promising for traffic target detection.