A Point-Cloud Data Analysis Framework for Early Deformation Detection

Frost heave and thaw-related deformation threaten the performance of roads, runways, embankments, and buried utilities in cold regions, yet traditional inspections remain reactive and often miss early uplift. This study evaluates the capability of LiDAR (Light Detection and Ranging) to detect small-scale ground deformation and introduces a practical point-cloud framework designed for early-stage characterization. Controlled jack-lift experiments quantified handheld LiDAR accuracy at uplift increments of 6.35, 12.7, 19.05, and 25.4 mm, with corresponding detection accuracies of 88%, 86%, 93.33%, and 94%. Registration quality had a significant impact. The targetless scans showed slight misalignment, whereas spherical targets provided precise, repeatable outputs. Although the framework functions without baselines or targets, the highest fidelity is achieved when both are available and when baseline subtraction is applied. A combined global and local roughness strategy was established to differentiate shallow uplift, with global metrics demonstrating greater sensitivity during the initial stages of deformation. The study enhances LiDAR-based (handheld for small scale; mobile and terrestrial for large-scale) deformation monitoring by quantifying accuracy under controlled uplift, formalizing a robust registration and change-detection framework, establishing a practical thresholding strategy, and identifying an optimal analysis grid. Grid-resolution analysis identified a 5 × 6 grid size (3.33% of the scan area) as the optimal balance between computational cost and accuracy. These results support earlier detection of ground deformation.