Real-Time YOLOv8 Detection and Particle Swarm Optimization Path Planning for COLREGs-Aware Collision Avoidance in Smart-Port Unmanned Surface Vehicle Operations

Abstract

Automation and digital transformation are being applied and widely implemented and advancing rapidly at seaports worldwide. With rapid growth, the number of ships entering and leaving the port is increasing, causing significant pollution in the port ecosystem. Therefore, regular monitoring and safety inspections at seaports need to be optimized and modern Artificial Intelligence (AI) technologies applied. Unmanned Surface Vehicles (USVs) are used for monitoring environmental quality with surveying water pollution at ports. However, current methods are often decoupled from object recognition and planning, and do not incorporate safety constraints. In this study, an object recognition method for seaports is proposed, utilizing the YOLOv8 deep learning model for multi-layer detection. Safe-distance calculations and safe distances and classifying situations are based on COLREGs. The Particle Swarm Optimization (PSO) selects the optimal route and generates waypoints (W) in USVs navigation. The experimental model is implemented in Unity 3D simulation software, and the controller is written in MATLAB. The research results show that mAP50 = 0.90 and 39 FPS demonstrate exemplary performance in the simulated smart-port seaport environment.