LiDAR-based ambient detection for surface vehicles in conurbations*

For autonomous vessels, it is important to have an accurate picture of static obstacles in a wide area to give surface vehicles enough time and space in case they need to stop or change direction. To share the position and outline of objects detected with LiDAR sensors, convex polygons are a compact method. In this paper, a voxel filter is proposed for the fusion of multiple LiDARs. The filter also takes into account previous measurements to stabilise the point cloud. In addition, the outlines of more distant objects are more accurately captured as they are scanned at different points in different measurements as the boat moves. The point clouds are segmented and concave hulls are calculated for the individual clusters, as these can maintain the contours with an adjustable accuracy. The concave polygons are then decomposed into convex ones. The result is a 2D map consisting of convex polygons. Furthermore, the individual convex polygons can be assigned to the individual static objects, which enables the integration of data from classification algorithms. These can then be used for subsequent algorithms, e.g. to improve the prediction of the path of other ships or for their own path planning. The presented method was successfully tested with an experimental vessel equipped with 2 LiDAR sensors.