Traffic Monitoring from the Perspective of an Unmanned Aerial Vehicle

Abstract

The paper is focused on the development of the experimental web-based solution for image processing from the perspective of an Unmanned Aerial Vehicle (UAV). Specifically, the research is carried out as part of the broader study on drone utilization in traffic at the Technical University of Kosice. This contribution explores the possibility of using the UAV as a tool to detect the temporal state of the traffic in multiple locations. Road traffic analysis is enabled through the detection of vehicles from the user-defined region of interest (ROI). Its content then serves as the input for motion detection, followed by the detection of vehicles using the YOLOv4 model. Detection of other types of objects is possible, thus making the system more universal. The vehicle is tracked after recognition in two consecutive frames. The tracking algorithm is based on the calculation of the Euclidean distance and the intersection of the rectangles. The experimental verification yields lower hardware requirements for CPU and GPU by about two FPS when using optimization techniques, such as ROI or reference dimensions of objects. The accuracy of detection and the subsequent tracking of cars reaches almost 100% while providing accurate trajectory determination.

Keywords

• computer vision
• deep learning
• object tracking
• unmanned aerial vehicle
• YOLO