Scholarlink

3D models are currently being utilized for the inspection, remote operation, and documentation of underground spaces such as tunnels. Research has been conducted for some time on creating 3D models of tunnels, involving the generation of high-precision point cloud data using LiDAR, in addition to employing photogrammetry as a cost-effective method for model creation. Recently, new photogrammetry techniques such as Neural Radiance Fields and 3D Gaussian splatting have been developed in computer science field. These methods can produce models of higher accuracy compared to previous photogrammetry techniques. In this study, a tunnel in Japan was captured using 360° cameras, stereo cameras, and LiDAR, documenting various shooting conditions such as illumination and time required for each capture method. From the video captured by a 360° camera, 3D models of the tunnel were developed employing techniques such as Structure-from-Motion Multi-View Stereo, Neural Radiance Fields, and 3D Gaussian splatting. Additionally, stereo matching techniques were used with stereo cameras to construct 3D models, and LiDAR data were utilized to create point cloud models. The accuracy and required time for each model were then compared.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Investigation of Image Observation Techniques and 3D Modeling Methods for Efficient Tunneling Operation

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yuta Kagaya,
  • Naohiro Ohtsuka,
  • Masahiro Katayama,
  • Shigetaka Ishihama,
  • Koichi Aoki,
  • Yoko Ohtomo,
  • Youhei Kawamura

摘要

3D models are currently being utilized for the inspection, remote operation, and documentation of underground spaces such as tunnels. Research has been conducted for some time on creating 3D models of tunnels, involving the generation of high-precision point cloud data using LiDAR, in addition to employing photogrammetry as a cost-effective method for model creation. Recently, new photogrammetry techniques such as Neural Radiance Fields and 3D Gaussian splatting have been developed in computer science field. These methods can produce models of higher accuracy compared to previous photogrammetry techniques. In this study, a tunnel in Japan was captured using 360° cameras, stereo cameras, and LiDAR, documenting various shooting conditions such as illumination and time required for each capture method. From the video captured by a 360° camera, 3D models of the tunnel were developed employing techniques such as Structure-from-Motion Multi-View Stereo, Neural Radiance Fields, and 3D Gaussian splatting. Additionally, stereo matching techniques were used with stereo cameras to construct 3D models, and LiDAR data were utilized to create point cloud models. The accuracy and required time for each model were then compared.