10th International Congress on Information and Communication Technology

Authors - Cynthia Sherin B, Poovammal E
Abstract - Vehicle detection and classification has a key role in evolution of intelligent transportation system. Accuracy in the detection enhances the efficiency of intelligent traffic monitoring systems. The paper shows a comparative study on the performance of YOLOv8 and YOLOv10 detection models on the novel Vehicle Identification aNd Classification (VINC) dataset introduced. They detect multi-class vehicles such as cars, trucks, buses, bicycles and bikes. The achievements of each model are assessed using precision, recall, F1 score and confusion matrices. The experimental results demonstrates the supremacy of YOLOv10 in detecting very small and more complex vehicle structures in the traffic scenario than YOLOv8. Alternatively, YOLOv8 also exhibited equivalent detection accuracy in detecting large vehicles like buses and trucks, by capturing the minute variations in the processed features. The detection models achieve precision of 97.2% and 93.6% for YOLOv10 and YOLOv8 respectively. YOLOv10 achieves high recall rate and F1 score of 92.4% and 81.4% respectively. Thus, the detection performance of these vehicles expresses the robust characteristics of both the YOLO versions. This research paper delineates the merits and drawbacks of these two versions on real-time circumstances, thereby creating faster and precise detection models.