RandLA-Net-Assisted Scan-vs-BIM Framework with Adaptive Parameter Optimization for Construction Progress Monitoring

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.