Publication

Abstract :

Naval mines pose persistent threats to maritime security, targeting submarines, warships, and critical infrastructure. While traditional detection methods employ side-scan and multibeam sonars, their effectiveness is limited by acoustic interference, and insufficient resolution, particularly at short ranges. This letter presents a real-time vision-based system for detecting and tracking visually similar naval mine-like targets of varying dimensions using a tethered remotely operated vehicle (ROV) equipped with a high-resolution optical vision sensor and an NVIDIA Jetson AGX Xavier edge AI module. Experiments conducted in a controlled swimming pool under daylight and low-light conditions demonstrate robust detection capabilities. The system achieves up to 94% accuracy in daylight and a minimum of 79% in low-light conditions, reliably distinguishing between two similar targets with same color and shape but varying in dimensions. The proposed approach highlights the potential of optical vision sensor in conjunction with edge AI for underwater mine detection in environments where traditional sonar-based methods face limitations.