Vision-Based Multi-detection and Tracking of Vehicles Using the Convolutional Neural Network Model YOLO

Abstract

Vehicle detection and tracking play a critical role in Intelligent Traffic Systems (ITS) for managing, identifying, and understanding the behaviour of vehicles on the road. In recent years, researchers have focused on developing computer vision-based approaches to improve the accuracy of vehicle detection and tracking. However, state-of-the-art approaches face challenges such as uncontrollable environmental conditions, occlusion of vehicles in a single frame, and high levels of noise, which can affect the accuracy of vehicle detection. In this study, we propose a computer vision-based approach to identify and track the movement of vehicles using the “YOU ONLY LOOK ONCE” (YOLO) version 7 Convolutional Neural Network (CNN) model. The YOLO model is fine-tuned to detect different types of vehicles, including cars, ambulances, buses, trucks, motorcycles, and bicycles. We utilized the Open Images Dataset from Google to test and validate our approach. Our experimental results demonstrate that the proposed approach achieves a higher detection accuracy of 81.28% mean Average Precision (mAP) compared to the state-of-the-art approaches. Furthermore, we tested our approach on the DAWN dataset to demonstrate its effectiveness as a multi-detector of vehicles in different weather conditions and online videos on the road for vehicle detection and tracking. Despite the challenges in the state-of-the-art computer vision-based ITS, our proposed approach demonstrates that it is possible to achieve higher accuracy using the YOLO version 7 CNN model. This study highlights the importance of developing accurate and efficient vehicle detection and tracking techniques to improve the operational functions of Intelligent Traffic Systems.