|本期目录/Table of Contents|

[1]尤桢杰,王家奎,熊 伦,等.一种基于神经网络隐式表达的室内建模改进方法[J].武汉工程大学学报,2025,47(02):202-209.[doi:10.19843/j.cnki.CN42-1779/TQ.202312029]
 YOU Zhenjie,WANG Jiakui,XIONG Lun,et al.An improved indoor modeling method based on neural network implicit representation[J].Journal of Wuhan Institute of Technology,2025,47(02):202-209.[doi:10.19843/j.cnki.CN42-1779/TQ.202312029]
点击复制

一种基于神经网络隐式表达的室内建模改进方法
(/HTML)
分享到:

《武汉工程大学学报》[ISSN:1674-2869/CN:42-1779/TQ]

卷:
47
期数:
2025年02期
页码:
202-209
栏目:
机电与信息工程
出版日期:
2025-05-09

文章信息/Info

Title:
An improved indoor modeling method based on neural network implicit representation
文章编号:
1674 - 2869(2025)02 - 0202 - 08
作者:
1. 光学信息与模式识别湖北省重点实验室,湖北 武汉 430205;
2. 武汉工程大学光电信息与能源工程学院、数理学院,湖北 武汉 430205;
3. 武汉唯理科技有限公司,湖北 武汉 430200
Author(s):
1. Hubei Key Laboratory of Optical Information and Pattern Recognition, Wuhan 430205,China;
2. School of Optical Information and Energy Engineering,School of Mathematics and Physics,Wuhan Institute of Technology,Wuhan 430205, China;
3. Wuhan Veily Technology Co., Ltd,Wuhan 430200,China
关键词:
Keywords:
分类号:
TP391
DOI:
10.19843/j.cnki.CN42-1779/TQ.202312029
文献标志码:
A
摘要:
本文提出一种基于神经网络隐式表达的室内三维重建改进方法,旨在解决仅使用彩色图像进行室内重建时效果不佳的问题。具体步骤包括:通过经典三维重建软件COLMAP获取相机参数和稀疏点云,利用线性特征的映射与定位工具箱LIMAP获取三维线段模型;通过深度估计和法向量估计获取深度信息和法向量信息;最终将深度信息、法向量信息和稀疏点云作为先验信息输入神经辐射场,以提高重建精度。在Scannet公开数据集上的实验结果显示,引入先验信息显著提升了重建效果,F分数达到0.70,峰值信噪比约为24 dB。在Scannet公开数据集和自建数据集上的实验结果表明,该方法有效解决了弱纹理区域的重建缺陷,显著提升了细节重建效果,对虚拟现实和数字建筑应用具有重要意义。
Abstract:
This paper proposes an improved indoor 3D reconstruction method based on neural network implicit representation, aiming to solve the problem of poor reconstruction effect when only using color images for indoor reconstruction. The specific steps include: obtaining camera parameters and sparse point clouds through the classic 3D reconstruction software COLMAP, and using the Linear Feature Mapping and Localization Toolbox LIMAP to obtain a 3D line segment model; obtaining depth information and normal vector information through depth estimation and normal vector estimation; finally, inputting depth information, normal vector information, and sparse point clouds as prior information into the neural radiance field to improve the reconstruction accuracy. Experimental results on the Scannet public dataset showed that introducing prior information significantly improves the reconstruction effect, with an F-score of 0.70 and a peak signal-to-noise ratio of approximately 24 dB. The experimental results on the Scannet public dataset and the self-built dataset demonstrate that this method effectively solves the reconstruction defect in weak-texture areas, significantly improving the detail reconstruction effect, hence has substantial promise for applications in virtual reality and digital architecture.

参考文献/References:

[1] 刘学人,刘康,王安妮,等. 视觉SLAM技术在室内三维重建中的应用[J]. 测绘技术装备,2023,25(4): 77-80.
[2] 董玲洁,代广红,冀松. 基于3D激光投影的室内场景三维重建模型[J]. 应用激光,2022,42(7):125-132.
[3] 王玮琦, 游雄, 杨剑,等. 一种改进匹配点对选取策略的ElasticFusion室内三维重建算法[J]. 武汉大学学报(信息科学版), 2020, 45(9): 1469-1477.
[4] 危双丰, 刘振彬, 赵江洪,等. SLAM室内三维重建技术综述[J]. 测绘科学, 2018, 43(7): 15-26.
[5] 王旒军, 陈家斌, 余欢,等. RGB-D SLAM综述[J]. 导航定位与授时, 2017, 4(6):9-18.
[6] 梅峰, 刘京, 李淳芃,等. 基于RGB-D深度相机的室内场景重建[J]. 中国图象图形学报, 2015, 20(10): 1366-1373.
[7] SCH?NBERGER J L, FRAHM J M. Structure-from-motion revisited[C]//2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2016: 4104-4113.
[8] MILDENHALL B, SRINIVASAN P P, TANCIK M, et al. NeRF: representing scenes as neural radiance fields for view synthesis[J]. Communications of the ACM, 2021, 65(1): 99-106.
[9] YARIV L, GU J T, KASTEN Y, et al. Volume rendering of neural implicit surfaces[C]//Advances in Neural Information Processing Systems. [S.l.]: Neural Information Processing Systems Foundation,2021:4805-4815.
[10] WANG P, LIU L J, LIU Y, et al. NeuS: learning neural implicit surfaces by volume rendering for multi-view reconstruction[Z/OL].(2023-02-01)[2023-12-29]. https://doi.org/10.48550/arXiv. 2106. 10689.
[11] FU Q C, XU Q S, ONG Y S, et al. Geo-Neus: geometry-consistent neural implicit surfaces learning for multi-view reconstruction[C]//Advance in Neural Information Processing Systems. [S.l.]: Neural Information Processing Systems Foundation,2022: 3403-3416.
[12] LIU S H, YU Y F, PAUTRAT R, et al. 3D line mapping revisited[C]//2023 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2023: 21445-21455.
[13] 陈起凤,刘军,李威,等. 三维重建中线段匹配方法的研究[J]. 武汉工程大学学报,2018,40(4):446-450.
[14] PARK J J, FLORENCE P, STRAUB J, et al. DeepSDF: learning continuous signed distance functions for shape representation[C]//2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2019: 165-174.
[15] 郑游, 王磊, 杨紫文. 基于多尺度深度图自适应融合的单目深度估计[J]. 武汉工程大学学报, 2024, 46(1): 85-90.
[16] EFTEKHAR A, SAX A, MALIK J, et al. Omnidata: a scalable pipeline for making multi-task mid-level vision datasets from 3D scans[C]//2021 IEEE/CVF International Conference on Computer Vision (ICCV). Piscataway: IEEE, 2021: 10766-10776.
[17] DAI A, CHANG A X, SAVVA M, et al. ScanNet: richly-annotated 3D reconstructions of indoor scenes[C]//2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Piscataway: IEEE, 2017: 2432-2443.

相似文献/References:

备注/Memo

备注/Memo:
收稿日期:2023-12-29
基金项目:国家自然科学基金(91963207)
作者简介:尤桢杰,硕士研究生。Email:854057029@qq.com
*通信作者:余子洋,博士,讲师。 Email:tommyu91@163.com
引文格式:尤桢杰,王家奎,熊伦,等. 一种基于神经网络隐式表达的室内建模改进方法[J]. 武汉工程大学学报,2025,47(2):202-209.
更新日期/Last Update: 2025-05-08