Automated Quality Assessment for Construction Progress Using Multi-View 3D Reconstruction and BIM Alignment

The construction industry increasingly relies on digital twins and building information modeling (BIM) to streamline project management; however, the discrepancy between the as-designed model and the as-built state remains a critical source of cost overruns and structural failures. This paper proposes a robust, automated framework for construction quality assessment (CQA) and progress monitoring by integrating multi-view stereo (MVS) 3D reconstruction with semantic BIM alignment. Unlike traditional manual inspections or expensive terrestrial laser scanning (TLS) methods, our approach utilizes cost-effective unmanned aerial vehicle (UAV) imagery to generate high-density point clouds. We introduce a novel semantic-aware registration pipeline that leverages deep learning-based segmentation to align noisy reconstruction data with pristine BIM geometries, effectively mitigating the challenges posed by dynamic construction environments and varying lighting conditions. By calculating the Euclidean deviations between the reconstructed mesh and the BIM elements, the system automatically classifies construction components into accurate, deviated, or missing categories. Experimental validation on complex high-rise construction datasets demonstrates that the proposed method achieves sub-centimeter accuracy in deviation detection and significantly outperforms conventional Iterative Closest Point (ICP) variations in alignment robustness. The results suggest that this pipeline can substantially reduce inspection latency and facilitate proactive decision-making in large-scale infrastructure projects.