Application of UAV 3D Point Cloud Data in Highway Slope Disaster Investigation

The safety and stability of highway slopes have always been critical concerns in highway management and maintenance. However, mountain road slopes, characterized by their small scale, large quantity, and complex geological conditions, pose challenges for effective deformation detection, putting significant pressure on highway management and maintenance. To address the difficulty in identifying hidden hazards of highway slopes in mountainous areas, this study proposes the use of unmanned aerial vehicle (UAV) closed-range photogrammetry technology for slope inspection. The M3C2 (Multiscale Model-to-Model Cloud Comparison) algorithm is employed to accurately calculate differences in multi-phase three-dimensional point cloud data, enabling the detection of overall slope deformation with centimeter-level precision. Field verification conducted on a highway slope in Qingtian County, Zhejiang Province, demonstrated the applicability of UAV 3D point cloud data in investigating highway slope hazards, thereby enhancing inspection efficiency and quality and providing technical support for highway management and maintenance.

hazards of highway slopes in mountainous areas, this study proposes the use of unmanned aerial vehicle (UAV) closed-range photogrammetry technology for slope inspection.The M3C2 (Multiscale Model-to-Model Cloud Comparison) algorithm is employed to accurately calculate differences in multi-phase three-dimensional point cloud data, enabling the detection of overall slope deformation with centimeter-level precision.Field verification conducted on a highway slope in Qingtian County, Zhejiang Province, demonstrated the applicability of UAV 3D point cloud data in investigating highway slope hazards, thereby enhancing inspection efficiency and quality and providing technical support for highway management and maintenance.

Introduction
China's vast mountainous regions with complex geological structures experience a high frequency of geological disasters.With the rapid economic development, the mileage of mountain roads has surged, exacerbating challenges in highway slope management and maintenance due to their sheer number, small scale, and linear distribution along highways [1] .During the "14th Five-Year Plan" period, thousands of kilometers of roads are planned for construction or expansion in Zhejiang Province, leading to a proliferation of mountain road slopes.Monitoring the stability of these slopes and ensuring safe operation is imperative for highway maintenance departments [2] .
In response to the mounting pressure of slope management and maintenance along highways in mountainous regions, enhancing the efficiency and accuracy of highway slope safety inspections is paramount.Presently, manual and automatic monitoring are the primary methods for mountain road slope safety monitoring.However, due to the high cost of equipment, monitoring coverage remains low, with most slopes inspected manually.Manual detection suffers from low efficiency and accuracy due to subjective factors.Thus, developing new deformation identification technologies suitable for mountain highway slopes is crucial for timely detection of abnormal slope deformations, obtaining deformation characteristics in advance, and undertaking necessary monitoring or protection measures to reduce disaster losses and maintain highway safety [3] .
In recent years, advancements in remote sensing technology, such as synthetic aperture radar interferometry (InSAR) and UAV close-range photogrammetry, have been increasingly applied in geological hazard identification [4] .While InSAR technology offers wide coverage and high accuracy, its effectiveness is limited by satellite revisit cycles, which can range from days to months.Moreover, the complexity of processing satellite remote-sensing image data hampers its widespread application in daily highway maintenance.Conversely, UAV closerange photogrammetry technology is terrain-independent and easy to operate, making significant strides in geological disaster investigation [6,7] .
This study introduces a deformation identification technology for highway slopes based on three-dimensional point clouds.UAV close-range photogrammetry technology is employed to reconstruct highway slopes in 3D, yielding multiphase 3D point cloud data.The M3C2 algorithm registers the multi-phase 3D point cloud data, enabling the derivation of time-series deformation data for highway slopes.The proposed method was applied and validated on an actual highway slope.This technology promises to enhance the efficiency and accuracy of highway slope detection, thus holding considerable significance for highway slope maintenance.

M3C2 algorism
UAV close-range photogrammetry represents an emerging technology stemming from advancements in digital imaging and photogrammetry.It integrates computer technology, digital image processing, impact matching, pattern recognition, and other interdisciplinary theories and methods to capture high-resolution images through airborne camera photography.Post-processing involves deriving the shape, size, position, characteristics, and interrelationships of objects.Essentially, it constructs three-dimensional space from twodimensional images, accurately capturing spatial positions and surface texture information of relevant terrain and ground objects from processed three-dimensional reality models or point cloud data.
Presently, UAV close-range photogrammetry technology finds application primarily in engineering geology for geological environment exploration and analysis of rock mass structural planes.The Multiscale Model-to-Model Cloud Comparison (M3C2) algorithm is utilized to analyze and compute UAV-generated 3D point cloud data, yielding insights into the deformation characteristics of highway slopes.Unlike other point-cloud comparison algorithms, this method directly detects complex terrain changes in point clouds without grid generation, with change calculation being less influenced by spatial point density, surface roughness, and different sampling locations.

Figure 1. The calculation principle of M3C2 algorism
Given that appropriate core point cloud selection significantly enhances computational efficiency, downsampling the original data to obtain a core point cloud with low density and uniform distribution is imperative at the initial calculation stage.This approach substantially boosts computational efficiency and optimizes time complexity, thereby reducing computational time.
Following the selection of a suitable core point cloud, for any given core point Pcore within a radius of D/2, a plane is fitted with other point-cloud data in the vicinity, allowing the determination of the local normal vector N of the two-phase point cloud.Subsequently, the distance from all points within the core Pcore radius D/2 to the best fitting plane is recorded, with roughness σ(D) characterizing the standard deviation size.

( ) (
) where ai is the distance between the k th point and best-fitting plane within radius D/2, a is the average distance between the best-fitting plane and all point clouds within radius D/2, and M is the total number of point clouds distributed within D/2 radius.

4
Commencing from the fitting plane, where Pcore lies the normal direction and D/2 represents the radius, a cylinder intersecting the two-phase point cloud through Pcore and the normal vector N is identified.All point clouds n1 and n2 contained within the cylinder are then searched, and the average position of the point cloud column in the two phases along the normal vector is calculated.At this juncture, the average positions of the point clouds in the columns of the two phases are denoted as M1 and M2, respectively, with the difference between the two average positions representing the distance LM3C2, indicative of the change distance of the slope at the Pcore point.This process is iterated across the entire slope until all point clouds are traversed, thereby yielding the change in point clouds across the entire target region, i.e., the displacement of the slope surface, facilitating the identification of geological disasters such as landslides.During the calculation process, setting appropriate algorithm parameters directly impacts subsequent computations, with the projection radius d/2, normal vector radius D/2, and maximum calculation depth H representing the key parameters affecting calculation accuracy and efficiency.
As shown in Figure 1(d), upon calculating the point cloud distance, to estimate the measurement accuracy of local distance change and mitigate misjudgments in slope disaster identification stemming from various errors, it becomes imperative to further determine the spatial confidence interval, reducing the likelihood of geological disaster misidentification.Under the assumption that the errors of multiple measurements adhere to an independent Gaussian distribution, the Z-double-tailed difference significance test formula is employed to calculate the confidence interval with a confidence level above 0.95 when n1 and n2 ≥ 30.

( ) ( ) ( )
where n1 and n2 are the core points of the point clouds in the two phases under a projection radius of D/2 and REG represents the registration error of the two-phase point cloud.LOD95%(d) is the minimum variation distance of the confidence interval above 0.95 at the confidence level of the projection radius D/2.The calculation formula for the registration error REG of the twophase point cloud is where RMSE1 is the root-mean-square error of the reference point cloud and RMSE2 is the rootmean-square error of the contrast point cloud.When 4 <n1 and n2< 30, the T-double-tailed significance difference test formula is used to calculate the confidence interval, and the degrees of freedom DF can be calculated according to the following formula: In this formula, when n1 and n2 are less than 4, no confidence intervals are required.

Case study
The road slope under study is situated in Qingtian County, Zhejiang Province.Qingtian County, nestled in the low mountainous terrain of southern Zhejiang Province, comprises approximately 90% mountains and hills and 5% rivers and flats.Structurally, the region is positioned within the eastern side of the southern segment of the second-order first-order Xia Uplift Zone in Neocathaysia.Notably, rigid rocks dominate the area, characterized by faulted structural traces and gentle, undeveloped corrugations.The terrain inclines from northwest and southwest to southeast, featuring numerous small basins and the pronounced incision of large streams and rivers.Consequently, Qingtian County grapples with various geological hazards, including frequent occurrences of highway slope disasters.
According to historical disaster records, a slope on a road in Qingtian County collapsed on August 4, 2020, resulting in a landslide volume of approximately 4000 m 3 and subsequent disruption of road traffic.The condition of the highway slope is depicted in the figure below.The slope exhibited substantial weathering, with prominently developed rock joints and relative fragmentation, while the overlying surface comprised entirely of weathered residual material and Quaternary sedimentary soil.Presently, excavation of the upper slope section has concluded, with partial completion of anchor reinforcement measures.Analysis of the calculation outcomes reveals that the errors in three-dimensional point clouds all fall within 10 cm, with a trend of decreasing error corresponding to higher altitudes.This trend primarily stems from the heightened accuracy of aerial photography images at greater altitudes, thus yielding more precise calculation results.Concurrently, evaluation of the slope surface calculation results indicates overall stability and absence of deformation.
A review of historical disaster scenarios alongside analysis of the 3D reality model indicates that the entire highway slope remains stable.With no significant precipitation, typhoons, or other heavy rain disasters observed from August to October 2023, overall displacement or deformation is absent.Nevertheless, continuous monitoring of highway slopes is imperative, particularly during spring and summer, when large-scale precipitation events may occur.Measures should be implemented to avert further sliding and soil erosion.
By computing the discrepancy in 3D point cloud data, the entire deformation of the edge slope can be delineated.Consequently, in routine highway slope inspections, inspection frequency may range from weekly to monthly, contingent on regulatory requirements and slope risk levels.Maintenance personnel can efficiently survey slopes using drones, ascertain highway slope deformation characteristics through indoor three-dimensional point cloud generation computations, reevaluate slope risk levels based on deformation conditions, and adjust inspection frequencies accordingly.

Conclusion
By advancing UAV close-range photogrammetry technology for highway slopes, this study introduced a methodology to ascertain the displacement and deformation of such slopes through the analysis of differences in three-dimensional point cloud data.Application verification was conducted on a highway slope in Qingtian County, Zhejiang Province, demonstrating that UAVgenerated three-dimensional point cloud data exhibits accuracy at the centimeter level in identifying highway slope displacement and deformation.This approach efficiently captures the overall safety and stability status of highway slopes, thereby furnishing a foundational dataset and technical framework for subsequent soil and water conservation management efforts and comprehensive safety measures.

Figure 2 . 3 .
Figure 2. The filed condition of slope To ensure slope safety, UAV close-range photogrammetry was conducted in August and October 2023.The route-planning parameters are illustrated in the figure below.Reconstruction of a three-dimensional model was executed using two aerial images, yielding two sets of threedimensional point cloud data.To validate the accuracy of the 3D point cloud data computation results, an equipment box was positioned at the middle, foot, and curb of the slope, at elevations of 21, 37.5, and 17 cm, respectively, as depicted in the layout diagram.

Figure 4 .
Figure 4.The result of the calculation

Table 1 .
The errors of the boxes