Abstract：Aiming at the problem caused by the sparseness of outdoor laser point cloud scenes for semantic segmentation, a point cloud segmentation method based on deep learning is proposed. First, the laser point cloud scenes collected from five perspectives are processed, and parts of buildings with higher overlapping areas are selected in turn, and each group is registered by SAC-IA and ICP-based point cloud automatic registration methods. In order to construct a large outdoor scene with relatively uniform point cloud density, the public data set Semantic3D is used to train an outdoor point cloud segmentation model based on PointNet++, and the algorithm effect is verified on the test set. Finally, this model is used to segment the outdoor scene that has been constructed scenes and experimental results prove that point cloud scenes with multi-view registration can solve the problem of non-uniform sampling of point cloud scenes, so that the deep segmentation-based semantic segmentation model has a better recognition effect.
徐鹏,徐方勇,陈辉. 融合配准的多站室外大场景激光点云分割[J]. 计量学报, 2022, 43(3): 325-330.
XU Peng,XU Fang-yong,CHEN Hui. Large Scene Segmentation of Outdoor Laser Point Cloud Based on Fusion and Registration. Acta Metrologica Sinica, 2022, 43(3): 325-330.
［1］Qi R C, Su H, Mo K, et al. PointNet: Deep learning on point sets for 3d classification and segmentation ［C］//IEEE. Proceedings of IEEE Conference on Computer Visionand Pattern Recognition. Honolulu, USA, 2017: 77-85.
［2］Qi R C, Yi L, Su H, et al. PointNet++: Deep hierarchical feature learning on point sets in a metric space ［C］//Proceedings of Conference on Neural Information Processing Systems. Long Beach, USA, 2017: 5099-5108.
［3］Wang W, Yu R, Huang Q, et al. SGPN: Similarity group proposal network for 3d point cloud instance segmentation ［C］//IEEE. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA, 2018: 2569-2578.
［4］Uy M A, Lee G H. PointNetVLAD: Deep point cloud based retrieval for large-scale place recognition ［C］//IEEE. Proceedings of IEEE Conference on Computer Vision and Pattern Recognition. Salt Lake City, USA, 2018: 4470-4479.
［5］Arandjelovic R, Gronat P, Torii A, et al. NetVLAD: CNN architecture for weakly supervised place recognition ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2018, 40(6): 1437-1451.
［6］曹文意, 陈继民, 袁艳萍, 等. 基于多视图的三维模型采集系统的研制 ［J］. 计量学报, 2019, 40(6): 1000-1005.
Cao W Y, Chen J M, Yuan Y P, et al. Research on 3D Model Acquisition System Based on Multiple View ［J］. Acta Metrologica Sinica, 2019, 40(6): 1000-1005.
［7］陈辉, 杨剑, 黄晓铭, 等. 基于多视图立体视觉的沙堆三维尺寸测量研究 ［J］. 计量学报, 2019, 40(3): 403-408.
Chen H, Yang J, Huang X M, et al. Research on 3D Measurement of Sand Pile Based on Multi View Stereo Vision ［J］. Acta Metrologica Sinica, 2019, 40(3): 403-408.
［8］葛晓天, 卢小平, 王玉鹏, 等. 多测站激光点云数据的配准方法 ［J］. 测绘通报, 2010 (11): 15-17.
Ge X T, Lu X P, Wang Y P, et al. Registration Method of Laser Point Cloud Data in Multi-stations ［J］. Bulletin of Surveying and Mapping, 2010(11): 15-17.
［9］邵正伟, 席平. 基于八叉树编码的点云数据精简方法 ［J］. 工程图学学报, 2010, 31(4): 73-76.
Shao Z W, Xi P. Data Reduction for Point Cloud Using Octree Coding ［J］. Journal of Graphics, 2010, 31(4): 73-76.
［10］陈学伟, 朱耀麟, 武桐, 等. 基于SAC-IA和改进ICP算法的点云配准技术 ［J］. 西安工程大学学报, 2017, 31(3): 395-401.
Chen X W, Zhu Y L, Wu T, et al. The point cloud registration technology based on SAC-IA and improved IC P ［J］. Journal of Xian Polytechnic University, 2017, 31(3): 395-401.
［11］Schenker P S. Method for registration of 3-D shapes ［J］. IEEE Transactions on Pattern Analysis and Machine Intelligence, 1992, 14 (2): 239-256.
［12］Rusu R B, Blodow N, Beetz M. Fast point feature histograms (FPFH) for 3D registration ［C］//IEEE. Proceeding of IEEE conference on Robotics and Automation. Kobe, Japan, 2009: 3212-3217.
［13］Rusu R B, Blodow N, Marton Z, et al. Aligning point cloud views using persistent feature histograms ［C］// IEEE. Proceeding of IEEE conference on Intelligent Robots and Systems. Nice, France, 2008: 3384-3391.
2022, Vol. 43 
