A Hybrid Tree Point Cloud Compression Method for Complex City Scenes
Da Ai, Yurong Yang, Xiaoyang Zhang, Ying Liu · 2022 4th International Conference on Natural Language Processing (ICNLP) · 2022
Point cloud compression has become an important research issue of computer vision. A lossy point cloud compression method with hybrid tree is proposed to cope with the problems of feature loss and high bit-rates of complex 3D surfaces in existing techniques with tree structure. In complex city scenes, due to surface differences between various regions of the LiDAR point cloud, a partitioning module based on Euclidean clustering is added obtain the different partitions of point cloud, and then the number of partitions is reduced by Gaussian curvature and mean curvature. In the hybrid tree compression module, the minimum partition bounding box is employed to select the tree structure and hyper-parameters, and obtain the optimal bit rates allocation scheme. Experimental results reveal that the proposed method can effectively reduce the coding redundancy of point cloud and have better rate distortion performance while ensuring compression quality in complex city scenes.