AI-Powered Monocular Perception for Collision Avoidance and Resilient Navigation in Unmanned Surface Vessels

This paper introduces a robust, AI-driven monocular perception framework for Unmanned Surface Vessels (USVs). We propose a deep learning–based object detection pipeline optimized for real-time vessel recognition and collision avoidance using lightweight, single-camera vision systems. The primary contribution of this work is the first systematic evaluation of YOLOv12’s robustness in monocular USV perception under simulated maritime challenges such as glare, motion blur, and varying illumination conditions. Our framework benchmarks four state-of-the-art object detectors—YOLOv8, YOLOv11, YOLOv12, and RF-DETR—using an augmented Sea Vessels Dataset to assess resilience and performance. Experimental results show that YOLOv12 achieves the best balance between precision (93.8%) and recall (78.8%), enabling safe and reliable operation of small autonomous vessels. These findings demonstrate that, with advanced AI models, monocular vision can deliver trustworthy situational awareness and decision support for next-generation connected and resilient maritime mobility systems.