Indoor scene 3D Mesh model reconstruction from point cloud with density variation using POCO

doi:10.13474/j.cnki.11-2246.2025.0202

Abstract

Abstract: To address the suboptimal performance of existing point cloud-based indoor 3D reconstruction methods when handling large-scale point cloud with significant density variations, we propose an improved indoor scene 3D Mesh model reconstruction method based on the POCO deep neural network, featuring enhanced training and reconstruction strategies. Firstly, we improve the training strategy by fine-tuning the original model using a small amount of simulated scene data and a very limited amount of real scene data. Secondly, we enhance the reconstruction strategy by introducing farthest point sampling and a strategy to ensure consistent bounding box scales. Finally, we restore the scale of the reconstructed 3D Mesh model. Experimental results demonstrate that our method outperforms the original POCO model in terms of reconstruction accuracy and model completeness, providing support for China's national 3D mapping program (3dRGLM).

Key words: POCO, 3D reconstruction, indoor scenes, point cloud with density variation, deep learning

CLC Number:

P237

SONG Peiyan, YE Qin, ZENG Liang, LUO Junqi, ZHANG Shuo, YIN Changjiang. Indoor scene 3D Mesh model reconstruction from point cloud with density variation using POCO[J]. Bulletin of Surveying and Mapping, 2025, 0(2): 7-12,47.

References

[1] 陈军,田海波,高崟,等. 实景三维中国的总体架构与主体技术[J/OL]. 测绘学报,2024(2024-04-18).https://kns.cnki.net/kcms/detail/11.2089.P.20240417.0946.002.html.
[2] 鲁一慧,魏国忠,宋禄楷,等. 多源点云优化的城市三维模型构建[J]. 测绘通报,2024(S1): 23-28.
[3] 杨泽鑫,程效军,丁琼,等. 面向室内场景点云的对象重建[J]. 测绘通报,2017(6): 45-48.
[4] XU Yusheng,STILLA U. Toward building and civil infrastructure reconstruction from point clouds: a review on data and key techniques[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing,2021,14: 2857-2885.
[5] 康志忠,杨俊涛. 室内实景三维重建技术综述[J]. 时空信息学报,2024,31(1): 1-10.
[6] 刘翔宇,王健,常清法,等. 改进贪婪投影三角化算法的激光点云快速三维重建[J]. 激光与红外,2022,52(5): 763-770.
[7] HOU Fei,WANG Chiyu,WANG Wencheng,et al. Iterative Poisson surface reconstruction (iPSR) for unoriented points[J]. ACM Transactions on Graphics,2022,41(4): 1-13.
[8] MESCHEDER L,OECHSLE M,NIEMEYER M,et al. Occupancy networks: learning 3D reconstruction in function space[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach: IEEE,2019: 4455-4465.
[9] PARK J J,FLORENCE P,STRAUB J,et al. DeepSDF: learning continuous signed distance functions for shape representation[C]//Proceedings of 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Long Beach: IEEE,2019: 165-174.
[10] BOULCH A,MARLET R. POCO: point convolution for surface reconstruction[C]//Proceedings of 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New Orleans: IEEE,2022: 6292-6304.
[11] WU Zhirong,SONG Shuran,KHOSLA A,et al. 3D ShapeNets: a deep representation for volumetric shapes[C]//Proceedings of 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Boston: IEEE,2015: 1912-1920.
[12] SHANMUGAM D,BLALOCK D,BALAKRISHNAN G,et al. Better aggregation in test-time augmentation[C]//Proceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV). Montreal: IEEE,2021: 1194-1203.
[13] BOULCH A,PUY G,MARLET R. FKAConv: feature-kernel alignment for point cloud convolution[M]//Lecture Notes in Computer Science. Cham: Springer International Publishing,2021: 381-399.
[14] HANDA A,WHELAN T,MCDONALD J,et al. A benchmark for RGB-D visual odometry,3D reconstruction and SLAM[C]//Proceedings of 2014 IEEE International Conference on Robotics and Automation (ICRA). Hong Kong: IEEE,2014: 1524-1531.
[15] 白静,司庆龙,秦飞巍. 基于卷积神经网络和投票机制的三维模型分类与检索[J]. 计算机辅助设计与图形学学报,2019,31(2): 303-314.