Road Damage Detection for Autonomous Driving Vehicles using YOLOv8 and Salp Swarm Algorithm

Abstract

Road accidents are one of the leading causes of death and serious injury in Malaysia, often resulting from human errors and poor road conditions. Autonomous vehicles aim to reduce accidents by mitigating human errors. Therefore, improving the road damage detection model in autonomous vehicles is crucial for enhancing their decision-making capabilities and reducing road accidents. Finding suitable sets of hyperparameters for this task is time-consuming. Consequently, this paper proposes a method to improve the detection accuracy of You Only Look Once version 8 (YOLOv8) using Salp Swarm Algorithm (SSA) for hyperparameter optimization, focusing on eight key parameters. The model is trained using the Czech data in Road Damage Dataset RDD2022 from the Crowdsensing-based Road Damage Detection Challenge (CRDDC’2022), with 80% of the data used for training and 20% for validation. The YOLOv8n model is trained with SSA on the RDD2022 dataset, specifically using data from India and China, to find the optimal parameters. The model is then retrained using the hyperparameters identified by SSA. The YOLOv8 models optimized using SSA are compared with the original YOLOv8 and other YOLO versions (YOLOv5, YOLOv9, and YOLOv10), demonstrating a 3.5% improvement in accuracy after hyperparameter optimization in detecting road damage.

Keywords

• hyperparameter optimization
• object detection
• salp swarm algorithm
• yolov8