Work

3D model-based reinforcement bar spacing measurement using self-supervised blind super-resolution and ultra-wideband localization

Year: 2026

Type: article

Abstract: In reinforced concrete construction, reinforcement bar spacing, diameter, and cover depth collectively govern structural stiffness and load-carrying performance. Among these parameters, bar spacing is typically the primary item verified during field quality control inspections. This study presents a three-dimensional model-based reinforcement bar spacing measurement system using a worker-worn handheld camera. However, images captured at real construction sites inevitably suffer from low resolution, blur, and noise due to dust, vibration, and confined spaces. Moreover, acquiring paired high-resolution ground-truth images under such conditions is practically infeasible, preventing the use of conventional supervised super-resolution methods. To address these limitations, a self-supervised blind super-resolution approach based on generative AI is adopted to enhance image quality without requiring paired high-resolution references. The proposed method effectively restores degraded images, even at previously unseen construction sites, achieving performance comparable to that of conventional supervised approaches and enabling reliable recovery of structural and textural details. Furthermore, ultra-wideband-based worker localization is fused with camera pose information to recover the metric scale of the three-dimensional reconstruction. Experimental results show that reinforcement bar spacing is measured with a mean absolute error of 29.13 mm and a relative error of 4.78% in large-scale real structures, suggesting that the proposed system provides acceptable measurement accuracy for verifying compliance with design specifications.

Source: Computer-Aided Civil and Infrastructure Engineering

DOI: 10.1016/j.cacaie.2026.100056

Authors: Seungbo Shim, Yeongjin Kim, Hyeongmo Park, Quang Ngoc Le

Institutions: Korea Institute of Civil Engineering and Building Technology, Incheon National University, Vietnam National University Ho Chi Minh City, Ho Chi Minh City University of Technology

Language: English

Topic: Infrastructure Maintenance and Monitoring

Subfield: Civil and Structural Engineering

Field: Engineering

Domain: Physical Sciences

SDG: Industry, innovation and infrastructure

Open Access status: hybrid

Estimated APC paid: $5,230