PVD: A Detection Framework for Parking Vehicles on Highways From Oblique UAV Images

Tracking parked vehicles on highways stands as a crucial aspect of traffic monitoring. Recent developments in intelligent transportation systems have led to the utilization of UAVs for overseeing parked vehicles using aerial images from a bird's-eye view (BEV). However, these methods face challenges in real-world settings due to restricted monitoring zones and potential breaches of safety protocols. This paper introduces a parking vehicle detection (PVD) framework designed to oversee parked vehicles on highways utilizing oblique images with significant camera tilt angles. Specifically, the PVD's perception module leverages advancements in deep learning to improve the accuracy of vehicle detection and lane segmentation. Additionally, an innovative BEV projection algorithm is introduced, which transforms oblique views into BEV based on camera parameters and pinhole model. Therefore, it can enable the PVD to measure the precise motion status of vehicles on the highway.Extensive experiments underscore the PVD's originality compared to existing approaches.