Development of Automatic Object Detection System for Autonomous Vehicles

Abstract

Autonomous vehicles (AVs) depend on advanced object detection systems to ensure safety and efficiency in navigation. Real-time detection of vehicles, pedestrians, and obstacles under diverse environmental conditions is vital for AV performance. This study proposes a real-time object detection system using the YOLOv5 architecture, a deep learning model recognized for its balance of speed and accuracy. The system aims to deliver precise and efficient detection to support safe navigation and decision-making in AVs. Real-time detection enhances situational awareness, obstacle avoidance, and overall driving safety, making it essential in autonomous systems. The literature review discusses conventional sensor fusion methods, such as integrating LIDAR and cameras, alongside modern machine learning approaches, especially deep learning and Convolutional Neural Networks (CNNs). YOLOv5 stands out for its capability in real-time processing, with proven success in detecting vehicles and pedestrians in dynamic environments. Despite these advantages, challenges like low-light conditions, occlusion, and variable weather still affect detection performance. The proposed system trains and fine-tunes YOLOv5 with diverse datasets to achieve reliable detection outcomes. Experimental results show that YOLOv5 maintains high accuracy and faster processing compared to traditional methods. This demonstrates its potential as a robust real-time detection framework for AVs. Future enhancements could involve incorporating multiple sensors and refining YOLOv5’s adaptability to complex driving scenarios