Online Obstacle Detection for USV based on Improved RANSAC Algorithm

Abstract

When an unmanned surface vehicle (USV) equiped with a LiDAR conducts obstacle detection, the swaying of the hull and the water splashes generated during navigation can cause disturbance and deviation in the scanned point cloud data, resulting in an increased rate of missed detection of static obstacles such as reefs and trees. This paper proposes an online obstacle detection algorithm for USV based on an improved Random Sample Consensus (RANSAC) algorithm. To address the large amount of point cloud data generated during the USV’s navigation process, a point cloud preprocessing based on voxel filtering is proposed to achieve denoising and compression of the original point cloud data while retaining its features. Considering that ground point cloud data will be disturbed during USV navigation, a RANSAC-based improved algorithm based on the grid projection method is designed, and ground segmentation is performed based on the results of static obstacle classification to generate a grid map. Clustering processing is performed using the grid clustering algorithm to obtain the detected obstacles and mark their location and size using bounding boxes. Finally, a trial run is conducted on a USV equipped with LiDAR, and the experimental results show that the proposed improved algorithm can reduce the missed detection rate and meet the real-time requirements of the algorithm, effectively improving the detection rate of nearby static obstacles.