Automatic Detection of Nearby Ships using Monocular Vision for Autonomous Navigation of USVs

Automatic detection and tracking of nearby ships are important capabilities for the autonomous operation of unmanned surface vehicles (USVs). This study focuses on achieving such capabilities in the framework of vision-based perception and sensor fusion. Reliable detection and tracking processes using a monocular camera are designed to automatically detect maneuvering targets with no prior information on the motion of targets. For a reliable trajectory estimation in low observability situations, the proposed vision-based approach uses bearing information in both the horizontal and vertical directions. In addition, the measurement by an onboard lidar is integrated into the vision-based tracking filter when the targets are in close range. The performance of the proposed method was assessed by field experiment data obtained in a real-sea environment to show the feasibility of the developed algorithms for the autonomous navigation of USVs.