基于3D高斯溅射在铁路三维可视化技术应用研究  

Research on the Application of 3D Gaussian Splatting in Railway Three-Dimensional Visualization Technology

作　　者：乔立贤 Qiao Lixian(Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China;Engineering Construction System Division,Beijing Jingwei Information Technology Co.,Ltd.,Beijing 100081,China;Key Laboratory of Railway Industry on BlM Software,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司电子计算技术研究所,北京100081 [2]北京经纬信息技术有限公司建设系统事业部,北京100081 [3]中国铁道科学研究院集团有限公司BIM软件铁路行业重点实验室,北京100081

出　　处：《铁道技术标准（中英文）》2024年第8期1-7,16,共8页Railway Technical Standard(Chinese & English)

基　　金：国家重点研发计划(2021YFB2600300);北京经纬信息技术有限公司科研项目(DZYF21-34)。

摘　　要：为提高铁路工程建设信息化水平,服务智慧铁路2.0,依托铁路工程管理平台深圳至汕头铁路2标信息化项目,研究基于3D高斯溅射在铁路的三维可视化技术。首先,制定基于3D高斯溅射在铁路三维可视化技术路线,通过数据层、区域化3D高斯溅射场、渲染层3层结构,采用无人机采集数据,3D高斯溅射机器学习,SuperSplat剪切、组装,最后应用Three.js-Gaussian Splats3D实现项目上的应用。其次,重点分析基于3D高斯溅射长大场景的区域化训练机制,铁路长大项目的区域化训练10 m缓冲区,SFM点云化处理,创建3D高斯椭球集,基于PyTorch的机器学习训练,3D高斯光栅化处理机制。最后,将3D高斯溅射在深圳至汕头铁路项目上试点应用,验证该技术路线的可行性,分析与传统倾斜摄影测量三维重建在渲染方面的优势。为铁路三维重建提供了一种新的技术路线。To improve the informatization level of railway engineering construction and serve the Smart Railway 2.0 initiative,the Shenzhen to Shantou railway section 2 informatization project,which relies on the railway engineering management platform,has researched 3D Gaussian splatting-based railway 3D visualization technology.Firstly,a technical route for railway 3D visualization based on 3D Gaussian splatting was established,utilizing a three-layer structure consisting of the data layer,regional 3D Gaussian splatting field,and rendering layer.This approach involves drone data collection,3D Gaussian splatting machine learning,SuperSplat shearing and assembling,and finally,implementing project applications using Three.js-Gaussian Splats3D.Secondly,the focus was on analyzing the regional training mechanism of 3D Gaussian splatting for long and large scenarios,with a 10m buffer zone for regional training of long and large railway projects;SFM point cloud processing;creation of 3D Gaussian ellipsoid sets;machine learning training based on PyTorch;and 3D Gaussian rasterization processing mechanism.Finally,a pilot application of 3D Gaussian splatting was conducted on the Shenzhen to Shantou railway project,verifying the feasibility of this technical route and analyzing its rendering advantages compared with traditional oblique photogrammetry 3D reconstruction.This provides a new technical route for railway 3D reconstruction.

关 键 词：3D高斯溅射 区域化 SFM 高斯椭球集 光栅化 

分 类 号：TP319[自动化与计算机技术—计算机软件与理论]