Visual and Infrared Detection and Ranging (VAIDAR) for Marine Navigational Hazards

The field of unmanned surface vessels (USVs) has made significant strides in autonomous navigation, yet accurate and reliable perception of the maritime environment, particularly in hazard detection, remains a challenge. Currently utilized technologies, such as radar and global positioning system (GPS), provide valuable data but EO/IR is underutilized due to complexities with processing. To address this problem, our research introduces Visual and Infrared Detection and Ranging (VAIDAR), a comprehensive solution that combines deep learning-based computer vision techniques with a physics-based algorithm for real-time EO/IR data processing that can offer two essential capabilities: hazard detection and hazard localization. VAIDAR has been found to provide a known safe distance of 2 nautical miles (NM) for a target 6.5 NM away, achieving an overall mean average precision (mAP) of 0.91. A true positive (TP) testing metric determined that the proposed method achieved a performance of 97.6% when using a contour overlap threshold of 50%. This indicates a robust congruity between the inference contours and the ground truth annotations across all the evaluated test frames. Together, this system and its findings will enhance USV autonomous navigation in various maritime environments.