Evaluating UAV Datasets for Vehicle Recognition in Mixed Traffic Conditions Using YOLO

This study evaluated the well-known UAV datasets for vehicle recognition in mixed traffic conditions using the state-of-the-art YOLOv8 object detection model. VisDrone2019-DET, VAID, and JATAYUv1.0 datasets were selected for their diverse vehicle class consideration, environmental conditions, and geographical representation. The performance of YOLOv8s on these datasets was assessed using metrics such as precision, recall, and mean Average Precision (mAP). Results showed significant variability in detection accuracy across datasets, highlighting the impact of dataset characteristics on model performance. JATAYUv1.0, tailored for heterogeneous traffic conditions, demonstrated superior detection for unique vehicle classes. VAID showed robust performance in intermediate altitude and varied camera angles, and VisDrone2019-DET excelled in long urban road stretches with varied lighting conditions. The study underscored the importance of dataset selection in UAV-based vehicle detection. It provided insights for researchers and practitioners in Intelligent Transportation Systems (ITS) to enhance traffic monitoring and safety applications.