Beyond Visible Light: Usage of Long Wave Infrared for Object Detection in Maritime Environment
Shailesh Nirgudkar, Paul Robinette
Abstract
Autonomous navigation of land vehicles has come a long way owing to the advances in scene parsing capabilities enabled by deep learning. Maritime environment shows an equally great promise in such autonomy and research specific to this domain is gaining rapid momentum. Extreme weather and light conditions are typical at seas and optical cameras commonly used on land face limitations in such environment. This paper proposes usage of Long Wave Infrared (LWIR) cameras to overcome these challenges. We evaluate the effectiveness of LWIR by running semantic segmentation against two deep learning architectures under normal and challenging conditions and then comparing their performance against their optical counterparts. Experimental results show that LWIR outperforms optical in such extreme weather and light conditions and will prove very effective in the maritime environment. Integrating LWIR with optical, radar and other sensors in the robotic navigation systems can improve the accuracy of object detection and avoidance in autonomous surface vehicles. The research also focuses on key areas where the algorithm can be enhanced as well as creation of a public maritime dataset of LWIR images as the next steps.