Road Train Detection and Decision Support Systems for Automated Vehicles using Deep Learning

This paper addresses the issue of road train detection and decision-making in real-world traffic scenarios. Road trains, comprising a convoy of vehicles, pose unique challenges for autonomous driving systems, but are also a major safety issue for civilian roads as road train crashes are often fatal. Our objective is to develop a decision-making framework to improve road safety for autonomous vehicles, which includes road train detection, the combination of camera and LiDAR sensory data, and lane detection. To tackle this problem, we employ the YOLO algorithm for object detection, specifically targeting road trains, and leveraging OpenCV for lane detection. After which, the developed framework is to be incorporated into the level 4 automated vehicle ZOE2 from QUT. Through testing various realworld cases, our system demonstrates its effectiveness in accurately detecting road trains and maintaining safe distances from them. During testing the YOLOv5 model was able to achieve a mean average precision (mAP) of approximately 0.74 for roadtrains and approximately 0.47 for long-vehicles. While the mAP @ 0.5 of the YOLOv8 model is approximately 0.81 for roadtrains and 0.8 for long-vehicles. Our study highlights the feasibility and adaptability of the proposed system for road train detection and decision-making.