基于点金字塔与极大团改进的残缺点云配准网络

Improved Partial Point Cloud Registration Network Based on Point Pyramid and Maximal Cliques

摘要

针对工业缺陷检测中,采集点云因零件受损、环境干扰等因素存在残缺,与标准点云配准时受几何结构差异影响,引发离群对应关系,导致配准精度下降的问题,提出一种基于点金字塔与极大团改进的残缺点云配准网络.首先,设计残缺补全模块生成虚拟点,该模块经构建的多分辨率动态图卷积提取点云的局部和全局特征后,添加点金字塔分形预测网络补全点云缺失区域,弥补缺失的特征信息;其次,在配准模块融合极大团算法过滤不兼容匹配对,优化匹配矩阵和位姿估计策略.在数据集ModelNet40上,对未知类别残缺点云配准实验的结果表明,该网络的旋转和平移平均绝对误差分别降至0.6507和0.0059,有效提升了离群对应关系配准的精度和鲁棒性;在数据集ESB上的实验结果进一步表明,该网络为工业缺陷检测提供了可靠的残缺点云配准方案.

Abstract

Aiming at the problem that in industrial defect detection, the collection of point cloud was incomplete due to factors such as component damage and environmental interference, structural discrepancies between incomplete and standard point clouds caused outlier correspondences during registration, resulting in a decrease in registration accuracy, we proposed an improved partial point cloud registration network based on point pyramid and maximal cliques. Firstly, we designed incomplete completion module to generate virtual points, the module extracted local and global features of point cloud through multi-resolution dynamic graph convolution, and added a point pyramid fractal prediction network to complete missing regions of the point cloud and compensated for the missing feature information. Secondly, the maximal cliques algorithm was integrated into the registration module to filter incompatible matching pairs, optimizing the matching matrix and pose estimation strategy. The results of unknown category partial point cloud registration on the ModelNet40 dataset show that the average absolute errors of rotation and translation of the proposed network are reduced to 0.6507 and 0.0059, respectively, effectively improving registration accuracy and robustness of outlier correspondences. The experiment on the ESB dataset further shows that the proposed network provides a reliable partial point cloud registration scheme for industrial defect detection.

关键词

点云配准 / 点金字塔 / 极大团 / 动态图卷积

Key words

point cloud registration / point pyramid / maximal cliques / dynamic graph convolution

参考文献

[1] 冯媛媛, 李朝奎, 刘松焕, 等. SSC-SeNet:一种融合点云与影像数据的露天矿区建筑物语义分割算法[J]. 吉林大学学报(地球科学版), 2025, 55(5):1757-1773.

[2] (Feng Y Y, Li C K, Liu S H, et al. SSC-SeNet: A Semantic Segmentation Algorithm for Buildings in Surface Mining Areas by Fusing Point Cloud and Image Data[J]. Journal of Jilin University(Earth Science Edition), 2025, 55(5): 1757-1773.)

[3] 王水仙, 邓朝晖, 葛吉民, 等. 基于工业三维检测的点云配准技术研究进展[J]. 金刚石与磨料磨具工程, 2023, 43(3):285-297.

[4] (Wang S X, Deng Z H, Ge J M, et al. Research Progress of Point Cloud Registration Technology Based on Industrial 3D Detection[J]. Diamond & Abrasives Engineering, 2023, 43(3): 285-297.)

[5] 王化明, 沈颖, 姜少华, 等. 基于点云的复杂工件机器人喷涂方法[J]. 江苏大学学报(自然科学版), 2026, 47(1):25-30.

[6] (Wang H M, Shen Y, Jiang S H, et al. Robot Spraying Method for Complex Workpiece Based on Point Cloud[J]. Journal of Jiangsu University(Natural Science Edition), 2026, 47(1): 25-30.)

[7] 万航, 吴晓群. 三维点云的配准方法综述[J]. 中国图象图形学报, 2025, 30(10):3255-3274.

[8] (Wan H, Wu X Q. Survey of 3D Point Cloud Registration[J]. Journal of Image and Graphics, 2025, 30(10): 3255-3274.)

[9] Wang Y, Solomon J M. Deep Closest Point: Learning Representations for Point Cloud Registration[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway, NJ: IEEE, 2019: 3523-3532.

[10] Wang Y, Solomon J M. Object DGCNN: 3D Object Detection Using Dynamic Graphs[J]. Advances in Neural Information Processing Systems, 2021, 34: 20745-20758.

[11] Zan G K, Wang Y, Gao P X. Improved DGCNN Based on Transformer for Point Cloud Segmentation[C]// International Symposium on Artificial Intelligence and Robotics. Berlin: Springer, 2023: 270-277.

[12] Wang Y, Slolmon J M. PRNet: Self-supervised Learning for Partial-to-Partial Registration[EB/OL]. (2019-10-27)[2025-04-18]. https://arxiv.org/abs/1910.12240.

[13] Deja K, Dubiński J, Nowak P, et al. End-to-End Sinkhorn Autoencoder with Noise Generator[J]. IEEE Access, 2020, 9: 7211-7219.

[14] Chaudhary L, Singh B. Gumbel-SoftMax Based Graph Convolution Network Approach for Community Detection[J]. International Journal of Information Technology, 2023, 15(6): 3063-3070.

[15] Xu H, Liu S C, Wang G F, et al. OMNet: Learning Overlapping Mask for Partial-to-Partial Point Cloud Registration[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision. Piscataway, NJ: IEEE, 2021: 3132-3141.

[16] Chen Z I, Chen H H, Gong L N, et al. UTOPIC: Uncertainty-Aware Overlap Prediction Network for Partial Point Cloud Registration[C]// Computer Graphics Forum. [S. l.]: Wiley, 2022: 87-98.

[17] 孔煜, 熊风光, 张志强, 等. 基于深度位置感知Transformer的低重叠点云配准[J]. 计算机科学, 2025, 52(5):199-211.

[18] (Kong Y, Xiong F G, Zhang Z Q, et al. Low Overlap Point Cloud Registration Method Based on Position-Aware Transformer[J]. Computer Science, 2025, 52(5): 199-211.)

[19] 徐守坤, 张路军, 石林, 等. 意图注意力引导的小样本3D点云目标检测[J]. 计算机工程, 2024, 50(12):288-295.

[20] (Xu S K, Zhang L J, Shi L, et al. Few-Shot 3D Point Cloud Object Detection Using Intention-Attention Guidance[J]. Computer Engineering, 2024, 50(12): 288-295.)

[21] Wang L H, Dai Q, Du T, et al. Novel Resampling Algorithms with Maximal Cliques for Class-Imbalance Problems[J]. Computers & Industrial Engineering, 2025, 199: 110754-1-110754-19.

[22] Wu Z K, Song S R, Khosla A, et al. 3D Shapenets: A Deep Representation for Volumetric Shapes[C]// Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. Piscataway, NJ: IEEE, 2015: 1912-1920.

[23] Wang M L, Chen G Y, Yang Y, et al. Point Tree Transformer for Point Cloud Registration[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2025, 35(7): 6756-6772.

[24] Zhu L K, Guan H N, Lin C W, et al. Leveraging Inlier Correspondences Proportion for Point Cloud Registration[EB/OL]. (2022-01-28)[2024-12-02]. https://arxiv.org/pdf/2201.12094.

[25] Liu Y Z, Liu Z Y. Low Overlapping Point Cloud Registration Using Mutual Prior Based Completion Network[J]. IEEE Transactions on Image Processing, 2024, 33: 4781-4795.

[26] Yew Z J, Lee G H. REGTR: End-to-End Point Cloud Correspondences with Transformers[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. Piscataway: IEEE, 2022: 6677-6686.

[27] Li J H, Zhang C H, Xu Z Y, et al. Iterative Distance-Aware Similarity Matrix Convolution with Mutual-Supervised Point Elimination for Efficient Point Cloud Registration[C]// Computer Vision-ECCV 2020: 16th European Conference. Berlin: Springer, 2020: 378-394.

[28] Huang X R, Gao X R, Li J L, et al. Clique-Like Point Cloud Registration: A Flexible Sampling Registration Method Based on Clique-Like for Low-Overlapping Point Cloud[J]. Sensors, 2024, 24(17): 5499-1-5499-17.

[29] Jayanti S, Kalyanaraman Y, Iyer N, et al. Developing an Engineering Shape Benchmark for CAD Models[J]. Computer-Aided Design, 2006, 38(9): 939-953.

基金资助

国家自然科学基金(61933007)

国家自然科学基金(61873058)

黑龙江省自然科学基金(LH2023F007)

国家重点研发计划项目子课题(2022YFC330160204)