Labor shortages in the construction industry have limited traditional manual progress monitoring, driving the demand for automated solutions. While many automated systems leverage the scan-vs-building information modeling (BIM) method, their robustness on complex construction sites remains a significant challenge. To address this, this study proposes a RandLA-Net-assisted framework with adaptive parameter optimization. The framework’s robustness is achieved through four key components: (1) a RandLA-Net-based clutter removal process for enhancing registration accuracy; (2) an oriented bounding box prealigned iterative closest point (OBBP-ICP) method for robust point cloud-to-BIM registration; (3) an adaptive coverage assessment (ACA) algorithm, which improves the resilience of element status recognition via an adaptive parameter; and (4) an integrated cost-level progress monitoring strategy. Validated on a live construction site in Hong Kong, the framework demonstrated an accuracy exceeding 99.23% in quantifying as-built concrete and curtain walls, with an average processing time of 47 min per scan. This framework significantly enhances the robustness and practicality of automated scan-vs-BIM progress monitoring.
Tang et al. (Thu,) studied this question.