Exploring the potential of backpack SLAM LiDAR for metro tunnel inspection: explainable modeling and optimization of point cloud quality

Researchers from Huazhong University of Science and Technology have developed an advanced modeling framework to enhance the quality of point cloud data from backpack SLAM LiDAR systems used in metro tunnel inspections. Published in Smart Construction, the study integrates statistical modeling with interpretable machine learning techniques to identify and optimize key inspection factors. Through comprehensive field experiments, the team demonstrated that inspection speed and scan density were crucial determinants of point cloud quality. This framework offers practical guidelines for improving data acquisition strategies, ensuring high-quality tunnel inspections, and promoting safer and more efficient metro operations.

Digital modeling of tunnels is essential for the effective management of urban metro infrastructure, providing critical data for structural recognition and risk assessment. Backpack LiDAR systems, with their portability and operational efficiency, have emerged as practical tools for tunnel inspections, especially in constrained metro environments. However, data quality issues such as instability, uneven density, and structural distortion frequently compromise the accuracy and reliability of these systems.

To address these challenges, researchers from Huazhong University of Science and Technology, led by Professor Cheng Zhou, have proposed a novel point cloud quality modeling framework combining statistical modeling and interpretable machine learning. Their study systematically evaluated five key inspection factors: movement speed, movement stability, scan density, tunnel curvature, and tunnel slope. A total of 12 field experiments were conducted in the Jiangji Tunnel section of Wuhan Metro Line 2, generating 360 point cloud samples. These samples were analyzed using four quality metrics: roughness, density, planarity, and sphericity.

The team applied the Skew-Normal distribution to statistically model the point cloud quality metrics, effectively capturing the complexity and variability of tunnel inspection scenarios. Subsequently, they constructed a CatBoost regression model enhanced by SHAP (SHapley Additive Explanations) for global and local interpretability, revealing clear causal relationships between inspection conditions and point cloud quality.

Results indicated that inspection speed and scan density consistently emerged as dominant factors influencing multiple aspects of point cloud quality. High movement speeds generally reduced data density and structural precision, whereas increased scan density significantly enhanced geometric completeness and detail accuracy. Interactions between factors, such as high speed combined with unstable movement or steep tunnel slopes, further exacerbated data quality issues, underlining the importance of carefully controlled operational conditions.

Additionally, the study’s interpretative analysis offered practical guidance on optimal operational speeds and scanning strategies, particularly highlighting moderate walking speeds and stable scanning paths as critical conditions for achieving high-quality data. This nuanced understanding enables field operators to effectively balance operational efficiency with data accuracy, enhancing inspection reliability.

This pioneering research provides an essential theoretical and practical foundation for optimizing backpack SLAM LiDAR systems in metro tunnel environments. The proposed modeling framework demonstrates significant engineering applicability and scalability, with potential extensions to broader infrastructure monitoring scenarios. Future research directions include exploring multi-source data integration and dynamic SLAM error modeling, aiming to further enhance the robustness and adaptability of tunnel inspection technologies.

This paper ” Exploring the potential of backpack SLAM LiDAR for metro tunnel inspection: explainable modeling and optimization of point cloud quality” was published in Smart Construction.
Qin W, Gao S, Zhou C. Exploring the potential of backpack SLAM LiDAR for metro tunnel inspection: explainable modeling and optimization of point cloud quality. Smart Constr. 2025(3):0017, https://doi.org/10.55092/sc20250017.