Multi-line LiDAR 3D construction environment modeling and BIM consistency update method for digital twins.

Area of Science:

• Digital Twins
• Construction Management
• LiDAR Technology
• Building Information Modeling (BIM)

Background:

• Current construction digital twins struggle with accurate, temporally stable as-built state representation from sequential LiDAR data.
• Existing methods like geometry-driven registration are brittle, while perception models lack temporal stability and BIM consistency.
• Limited state representations in digital twins hinder reliable and causal site updates.

Purpose of the Study:

• To develop a causal LiDAR-BIM digital twin (LiBiDT) for consistent state evolution through integrated spatio-temporal mapping, perception, and association.
• To enable online updating of construction digital twins using only past observations for a temporally stable, element-centric BIM state.
• To provide auditable construction management indicators, including completion, deviation summaries, and safety/compliance hazard flags.

Main Methods:

• LiBiDT integrates Spatio-Temporal Global Registration (STGR) with sensor priors, Multi-task Perception (MTP) for semantic and geometric understanding, and Scan-to-BIM Association and State Updating (S2B).
• S2B employs normal-guided refinement, bipartite matching with fused evidence (overlap, deviation, semantics), and Exponential Moving Average (EMA) for stabilization.
• A Decision and Risk Analytics (DRA) layer processes the BIM-aligned state into actionable management indicators.

Main Results:

• LiBiDT achieved an association F1 score of 0.837 and low Chamfer Distance (3.2 cm) / surface RMSE (3.5 cm) on the CV4AEC 2024 Scan-to-BIM benchmark.
• Significant gains were observed compared to a perception-replacement baseline, with competitive results against the strongest existing methods, particularly in temporal identity consistency (IDF1 0.854 → 0.861).
• Ablation studies confirmed the complementary benefits of evidence stabilization and multi-cue matching for accuracy and temporal consistency, demonstrating a favorable efficiency-accuracy trade-off.

Conclusions:

• The proposed LiBiDT model effectively addresses the challenges of creating temporally stable, BIM-consistent as-built states from streaming LiDAR data.
• The integration of spatio-temporal mapping, multi-task perception, and robust association mechanisms enables reliable online updating and advanced construction management analytics.
• LiBiDT offers a promising framework for improving progress verification, dimensional compliance, and proactive safety management in the Architecture, Engineering, and Construction (AEC) industry.