Simplification and Compression Method Based on Classified Sparse Sampling for LiDAR Point Cloud

As a key component of point cloud acquisition technology, LiDAR (Light Detection and Ranging) acts as an eye in autonomous driving. The vehicle-mounted LiDAR point cloud system is the prerequisite foundation for the vehicle to be able to perform environmental perception, path planning, navigation, and other behaviors. In practical applications, LiDAR point clouds have a wide distribution range and a large amount of data, which poses major challenges to existing transmission and storage technologies. Aiming at the point cloud from automotive LiDAR in autonomous driving, we propose a point cloud lossy compression framework based on range image index and classified sparse sampling. This paper takes the original data packet as the compression target and converts it into a 2D range image and a 3D point cloud based on its data characteristics. On this basis, we can segment 3D point clouds into ground point and non-ground point cloud by applying progressive morphological filters. Then, we perform de-redundant sparse processing at different intensities on the two. A sparse 2D range image is generated by referring to the range image index of the downsampling 3D point cloud and is represented as a more compact form of one-dimensional range vector combined with its corresponding occupancy image and Morton sorting. Finally, the image coding method is used to further compress the point cloud. Experimental results show that our proposed method is better than the existing algorithms, thereby effectively reducing the spatial redundancy of the point cloud and achieving higher reconstruction quality.