Scholarlink

<p>State-of-the-art methods for inspecting reinforced concrete (RC) structures employ a variety of sensing technologies, including LiDAR, Ground Penetrating Radar, vision-based systems, and photogrammetry. While these techniques have significantly enhanced the precision and reliability of rebar inspections, the potential of emerging technologies like Neural Radiance Fields (NeRFs) as cost-effective alternatives remains underexplored. This research integrates NeRF into our previously developed framework for processing point cloud data, which includes a Component Recognition Module using an adapted RandLA-Net model for semantic segmentation and a Dimensional Assessment Module for assessing rebar spacing. We compare the performance of NeRF with photogrammetry-based methods within this framework to evaluate its feasibility and effectiveness for rebar inspection. Our results demonstrate that NeRF offers significant advantages in handling larger surface areas, producing more complete point clouds, and achieving semantic segmentation performance comparable to photogrammetry. Specifically, NeRF slightly outperforms photogrammetry in rebar spacing accuracy, exhibiting lower mean absolute error and mean squared error, along with reduced variability in measurements. These findings suggest that NeRF is a promising, cost-effective alternative for rebar inspection, expanding the capabilities of RC structure assessment and offering new insights into the potential of advanced 3D reconstruction in construction quality assurance.</p>

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

Evaluating neural radiance fields for effective rebar inspection

  • 期刊论文

  • 发表日期：

    2026年2月11日

查看全文
  • Mohamed Abdelmadjid Talbi,
  • Salah Bouchama,
  • Mehdi Ammi

摘要

State-of-the-art methods for inspecting reinforced concrete (RC) structures employ a variety of sensing technologies, including LiDAR, Ground Penetrating Radar, vision-based systems, and photogrammetry. While these techniques have significantly enhanced the precision and reliability of rebar inspections, the potential of emerging technologies like Neural Radiance Fields (NeRFs) as cost-effective alternatives remains underexplored. This research integrates NeRF into our previously developed framework for processing point cloud data, which includes a Component Recognition Module using an adapted RandLA-Net model for semantic segmentation and a Dimensional Assessment Module for assessing rebar spacing. We compare the performance of NeRF with photogrammetry-based methods within this framework to evaluate its feasibility and effectiveness for rebar inspection. Our results demonstrate that NeRF offers significant advantages in handling larger surface areas, producing more complete point clouds, and achieving semantic segmentation performance comparable to photogrammetry. Specifically, NeRF slightly outperforms photogrammetry in rebar spacing accuracy, exhibiting lower mean absolute error and mean squared error, along with reduced variability in measurements. These findings suggest that NeRF is a promising, cost-effective alternative for rebar inspection, expanding the capabilities of RC structure assessment and offering new insights into the potential of advanced 3D reconstruction in construction quality assurance.