Real-Time Object Detection in Autonomous Vehicles Using Deep Learning

Abstract

Real-time object detection is a fundamental requirement in autonomous vehicle systems to ensure safe navigation, obstacle avoidance, and efficient decision-making in dynamic driving environments. This study focuses on the development of a deep learning-based framework for real-time object detection in autonomous vehicles, addressing challenges such as varying lighting conditions, occlusions, and high-speed object movement. The proposed methodology utilizes convolutional neural networks (CNNs) and state-of-the-art object detection architectures such as You Only Look Once (YOLOv5/YOLOv8) and Single Shot MultiBox Detector (SSD) to identify and classify objects including vehicles, pedestrians, traffic signs, and road obstacles. The system processes input data from onboard cameras and sensors, followed by image preprocessing techniques such as normalization, augmentation, and noise reduction to enhance detection accuracy. The deep learning model is trained on large-scale autonomous driving datasets to improve generalization across diverse road conditions. Performance evaluation is conducted using metrics such as mean average precision (mAP), inference speed (frames per second), precision, recall, and localization accuracy. The results demonstrate that YOLO-based models achieve superior real-time performance with high detection accuracy and low latency compared to traditional computer vision methods. The system effectively detects multiple objects simultaneously under complex urban driving scenarios while maintaining real-time processing capability essential for autonomous navigation. It is also observed that optimized lightweight deep learning architectures significantly improve computational efficiency, making them suitable for embedded automotive systems.