Optimization of Modeling Accuracy for Mobile Mapping Systems in Large-Scale Environments

Geospatial data acquisition technology has been widely integrated into geological and engineering geology research, significantly enhancing the spatial precision and structural integrity of topographical interpretations. In the era of high-performance computing, 3D geological modeling has emerged as a pivotal trend for engineering applications. However, the practical depth of these models is often constrained by challenges in accuracy and reliability, arising from varying data resolutions and the complexities of integrating multi-source information. These issues complicate model validation, particularly in large-scale or high-complexity engineering environments. As data collection methods become increasingly diverse, Simultaneous Localization and Mapping (SLAM) technology has revolutionized traditional surveying by offering superior operational flexibility and mobility. Unlike static terrestrial laser scanning (TLS), handheld or mobile LiDAR systems (MMS) can efficiently traverse indoor spaces, narrow urban corridors, and densely vegetated areas, facilitating the construction of comprehensive, blind-spot-free 3D spatial datasets. Despite these advantages, achieving and maintaining engineering-grade precision in GNSS-denied or signal-unstable environments remains a critical technical bottleneck. This study aims to investigate a robust workflow for large-scale field model construction using a "batch processing and stitching fusion" strategy. Using the National Taipei University of Technology (NTUT) campus as an experimental field, high-density point cloud data were collected using the mobile mapping system. The research methodology focuses on optimizing geometric fidelity by rigorously analyzing two key variables: first, a comparative evaluation of trajectory adjustment modes, specifically contrasting loop-closure correction with Post-Processed Kinematic (PPK) technology; and second, an assessment of how the quantity and spatial distribution of Ground Control Points (GCPs) influence the model’s global stability and absolute correctness.

The experimental results demonstrate that through optimized GCP deployment and refined trajectory adjustment, the absolute accuracy of the point cloud model can be maintained within an RMSE of 5 cm, with the relative accuracy on ground surfaces controlled within 2 cm. Furthermore, in the measurement of high-rise structures, the ghosting effect (layering) is restricted to within 4 cm at a 30-meter operational radius, while an average point spacing of 4 cm is maintained to ensure the geometric integrity of model details. These findings confirm that mobile LiDAR systems, when supported by optimized workflows, can meet the stringent precision requirements of engineering-grade projects while retaining high flexibility.

In conclusion, this research establishes a high-precision 3D digital foundation for the campus. This methodology is highly extensible to geological fields, including outcrop geometric measurement, quantitative analysis of landslide volumes, and structural surveys in GNSS-denied environments such as tunnels and caves.