基于非一致性稀疏采样的LiDAR点云压缩方法

针对自动驾驶车载LiDAR点云，本文提出一种基于形态学分割和非一致性稀疏采样的新型有损点云压缩框架。LiDAR点云先经过渐进式形态学滤波器分割为地面和非地面点云两部分，对两者进行不同强度的去冗余稀疏采样，之后将3D数据经球坐标变换映射为2D矩阵(表示为距离图像)，并通过占据图形式表示距离图像像素值是否存在。根据占据图的Morton 码排序，2D矩阵被表示为更加紧凑的1维距离向量。最后对占据图和距离向量利用图像编码方法进行压缩。实验结果表明，本文方法压缩性能明显优于点云压缩锚点，Google Draco方法；与MPEG TMC13方法相比，在较大bpp的情况下可以达到更高的重建质量，恰好适于精度要求高的自动驾驶应用场合。

LiDAR point cloud compression method based on non-uniform sparse sampling

For the LiDAR point cloud of autonomous driving vehicles, this paper proposes a new lossy point cloud compression framework based on morphological segmentation and non-uniform sparse sampling. The LiDAR point cloud is first divided into two parts of the ground and non-ground point clouds through a progressive morphological filter, and the two parts are de-redundant and sparsely down-sampled with different intensities. And then the 3D data is mapped into a 2D matrix (represented as a distance image) through spherical coordinates transform, and the occupancy map is introduced to indicate whether the pixel value of the distance image exists. According to the Morton code ordering of the occupancy map, the 2D matrix is represented as a more compact 1-dimensional distance vector. Finally, the occupancy map and distance vector are compressed using image coding methods individually. The experimental results show that the compression performance of the proposed method is significantly better than the point cloud compression anchor, Google Draco method; compared with the MPEG TMC13 method, higher reconstruction quality can be achieved in the case of a larger bpp, which is suitable for automatic driving applications with high precision requirements.