Advancing Autonomous Vehicle Perception: YOLO Algorithms and Custom Dataset Integration for the UAE Environment

This paper presents a study on the optimization of You Only Look Once (YOLO) object detection algorithms for enhancing autonomous vehicle (AV) perception systems, with a focus on the United Arab Emirates (UAE) driving conditions. Through evaluation of YOLO variants in simulated environments, YOLOv5m proved optimal by balanced detection accuracy and processing efficiency. The research involved creating a UAE-specific dataset to address regional scenarios, integrating real-time data processing in a virtual environment, and assessing algorithm performance across different conditions. Findings demonstrate the importance of varied environmental condition testing for object detection Algorithms, Furthermore the successful application of the custom UAE dataset underscores the necessity for regionally adapted training data in achieving high precision in object detection that is essential for navigation requirements of an AV.