Key points are not available for this paper at this time.
In modular construction, quality control is a crucial step in meeting quality requirements, leading to the completion of a project within the planned schedule and cost. Currently, quality inspection is visually performed by inspectors, which can be costly and unreliable. This study thus proposes a vision-based approach to off-site quality inspection that reconstructs (three-dimensional) 3D point clouds using a projector-camera system and computes the deviations between scans and virtual model to generate error maps for quality assessment. Particularly, an experimental investigation is carried out to evaluate operational conditions (e.g., distance, illumination, and the size of point cloud data) for an understanding of the practical application from a technical viewpoint. The results demonstrated that the 3D reconstruction performance has an inverse relationship with illumination levels and statistically significant differences between projection distances. It is also observed that using an optimal number of points for registration reduces the considerable computational cost with no negative impact on registration accuracy. In the experiment, for instance, 3D registration of a mock-up model (e.g., 300,000 points) using an optimal number of sampled points (e.g., 500 points in this study) produced an RMSE of 2.462 mm and took 4 s. The results also indicate that the method achieved an inspection accuracy within 7 mm in dimension errors and correct detection of missing parts. Therefore, this study allows for potentially reducing the time and effort involved in an off-site quality inspection by systematically and objectively recognizing quality issues.
Bae et al. (Tue,) studied this question.