3D point cloud segmentation, as one of the important technology of 3D scene understanding, has aroused people's widespread interest. And it has important research value and broad application prospect. The latest research progress of 3D point cloud segmentation technology based on deep learning is sorted out. Firstly, eight indoor and outdoor common datasets frequently utilized in 3D point cloud segmentation are introduced. Then, the existing semantic segmentation, instance segmntation and part segmentation mainly via deep learning are explained and analyzed in depth, and the effectiveness of some methods is compared baesd on quantitative data. Finally, we summarize the shortcomings of the existing methods from ten aspects, and put forward the work prospects pertinently.