Obstacle Risk Assessment for Unmanned Surface Vehicle Using Camera and Lidar

This research aims to enhance vessel capabilities in assessing dynamically moving obstacles, such as floating debris and other vessels, by leveraging data from cameras and 3D Lidar sensors. YOLOv8 is trained to detect obstacles, with a specific focus on ship vessels. Point cloud processing is employed to identify clustered point clouds and match them with bounding boxes. The position obtained from the previous step is then utilized to estimate velocity using the Kalman filter. The system provides real-time information on the closest distance of approach and the remaining time before potential collisions occur. We use the remaining time before potential collisions occur to determine whether the obstacle is approaching or receding from us.