Tracking and Speed Estimation of Ground Vehicles Using Aerial-view Videos

With the rapid technology development in autonomous navigation of Unmanned Aerial Vehicles (UAVs) and robust object detection based on deep neural networks, the field of traffic analysis through aerial video has attracted widespread attention. In this paper, we investigate the problems of ground vehicle tracking and speed estimation using aerial view videos. At the first stage, the vehicle detection is performed through the YOLOv3 network, which is the state-of-the-art object detector. Then, a tracking-by-detection method is designed to tracking the traffic vehicles. Furthermore, in order to estimate the vehicle speed in traffic while the UAV navigating in different heights, the least square algorithm is utilized to fit the measurement data and determine the power function mapping relationship between the vehicle pixel distance and the actual distance, which further improves the accuracy of speed estimation effectively.