A Real-Time Vehicle Detection System for ADAS in Autonomous Vehicles Using YOLOv11 Deep Neural Network on Embedded Edge Platforms

This paper presents a novel real-time vehicle detection system for autonomous vehicles, leveraging the advanced capabilities of the YOLOv11 deep neural network. The study addresses key challenges, including varying lighting conditions, occlusions, and the need to detect both traditional vehicle classes (cars, buses, trucks, motorcycles) and emerging ones, like e-scooters and emergency vehicles (fire engines, ambulances, and police vehicles). A custom dataset of 38,500 annotated images, encompassing diverse real-world traffic scenarios, was utilized to train and evaluate the model. The optimized YOLOv11 model was deployed on embedded platforms, specifically the NVIDIA Jetson Nano and Raspberry Pi 5, achieving a balance between high detection accuracy and low resource consumption. The system achieved a precision of 98%, recall of 95%, and an F1-score of 96.5%, with a mAP@50 of 98.1%. The hardware-specific optimizations improved the inference speed and efficiency. The results demonstrate that YOLOv11, combined with lightweight embedded systems, offers a scalable, real-time solution for integration into Advanced Driver Assistance Systems (ADAS) and autonomous vehicles, ensuring safer and more reliable navigation.