LiDAR Simulator for Autonomous Driving in Ocean Scenes

The publicly available point cloud datasets and simulation datasets are both based on urban scenarios. There is currently no marine scene point cloud dataset that leading to the perception algorithms applied to ships are difficult to develop due to a lack of data. Therefore, this paper proposes a LiDAR simulator applied to maritime scenarios generating point cloud data for autonomous driving in ocean scenes. The simulator consists of two parts: one is the modeling of maritime objects, and the other is the establishment of a LiDAR simulator. For maritime objects, the sea scenes include ships, obstacles and waves. The ship 3D mesh models adopt proportional modeling. And a portion of the ship 3D models are obtained through the restoration of point clouds collected in real world. Water contain bubbles can be captured by LiDAR sensor in the sea surface. Therefore, we use classic wave models to simulate three dimensional point clouds of ocean waves to generate wave point clouds. For the establishment of LiDAR simulator, firstly, we use the Bounding Volume Hierarchy (BVHTree) to segment the marine scene objects into the smallest elements enclosing an axis-aligned bounding box (AABB). Then the intersection points between the laser rays emitted by the simulator and AABBs are the rendering point clouds. In order to more realistically restore the simulator, noise and reflectivity are also considered. Finally, we utilize the LiDAR simulator to generate point clouds for ocean scenes and annotate collected point clouds.