Vision-Based Human Detection for Maritime Search and Rescue Unmanned Surface Vehicles

Abstract

Maritime search and rescue (SAR) operations rely heavily on manned crews, and the time required to organize and deploy personnel can significantly delay the location and rescue of individuals in distress. Unmanned Surface Vessels (USVs) are robots that operate autonomously or semi-autonomously on the surface of the water. They have been applied to an expanding range of scenarios, including mine countermeasures, water surveying, oceanographic research, and SAR. However, limitations in water-level vision systems, autonomous navigation, and reliable data transfer complicate the effectiveness of a SAR USV. For example, the lack of representative datasets that feature people in water further impedes the development and training of SAR-applicable machine vision. To fill a gap in the research, we created applicable datasets, developed, tested, and trained a vision system, and integrated that vision system into a fully autonomous USV.