A Deep Learning-Based Framework for Accurate Detection and Classification of On-Road Vehicles Using Improved YOLOv11

Vehicle detection and classification systems have significantly improved in recent years due to the developments in deep-learning-based frameworks for object detection. These systems have various applications in autonomous driving, intelligent transportation, traffic management, and urban planning. We propose a framework for accurately detecting and classifying on-road vehicles using a deep learning model called You Only Look Once (YOLOv11). Our study provides a comprehensive knowledge of model efficiency in vehicle detection and classification across nine classes: bicycle, bus, car, e-bike, jeep, motorcycle, tricycle, truck, and van. We tested the performance of the proposed improved YOLOv11 model and evaluated it using four performance matrices. The proposed improved YOLOv11 model achieved a precision of 96.5% and a recall of 96%, an F1 score of 77%, and an AUPRC of 82%. We also compared the proposed model performance with other versions of the YOLO series, as well as various traditional deep learning models to determine the effectiveness in vehicle detection and classification. The framework is a strong option for real-time traffic monitoring and autonomous driving applications, as the results show that it greatly increases precision and recall, especially in high-traffic situations.