Verification of Slope 3D Measurement Accuracy: Achieving cm-Level Accuracy with GNSS × Point Clouds

Introduction: Advanced Slope Management and Growing Demand for 3D Surveying

In civil engineering sites, slope management has increasingly required higher sophistication in terms of safety and quality assurance. Traditionally, slope geometry has been checked by surveys using total stations (TS) or levels, but these methods are limited in measurement coverage and point density and impose a heavy workload. Meanwhile, advances in 3D technologies such as laser scanners and drone photogrammetry have increased demand for three-dimensional surveying that captures terrain, including slopes, as surfaces. With the momentum of i-Construction promoted by the Ministry of Land, Infrastructure, Transport and Tourism, initiatives to use point cloud data for as-built management are spreading. However, when transitioning to new technologies, many express concern about whether accuracy can be maintained at conventional levels. This article verifies the accuracy of slope measurement using a modern approach that combines real-time kinematic positioning (RTK) with GNSS and smartphone 3D point cloud capture. After organizing the measurement requirements for slope surveying from a technical perspective, we explain the rationale for achieving cm-level accuracy (half-inch accuracy) using demonstration data and comparisons with conventional methods.

Technical Perspective: Slope Survey Items and Accuracy Requirements

When surveying slopes, several important items must be measured. First, the coordinates of the top of the slope (crest) and the bottom (toe) must be accurately determined because they affect property boundaries and interfaces with structures. Checking whether the slope gradient (inclination angle) matches the design is also a key point. For example, if the design calls for a gradient of 1:1.5, it must be confirmed that this gradient is maintained uniformly along the length and that there are no sudden flattenings or steepenings. Surface irregularities and finishing thickness of the slope are also targets for measurement. In particular, when the slope is protected with sprayed concrete or vegetation mats, it is necessary to confirm that the surface is smooth and the specified thickness is secured.

The required accuracy levels for measuring such slope geometry must also be understood. In general, tolerances for as-built dimensions in civil works are set on the order of several centimeters. For example, the finished position of a slope often needs to be within approximately ±5 cm (±2.0 in) of the design, and inspections will check whether this is the case. In other words, surveying must be capable of reliably detecting errors of a few centimeters. Single-point observations with a total station can achieve millimeter-level accuracy, but the number of points obtainable at once is limited. Therefore, point cloud surveying, which can acquire dense surface data, becomes effective. The Ministry of Land, Infrastructure, Transport and Tourism’s "3D As-Built Management Guidelines (draft)" sets a measurement density criterion for earthworks using TLS (terrestrial laser scanner) of at least 100 points per 1 m² (one point per 0.01 m²). This aims to cover the entire slope with points so that small bumps and hollows are not missed and the geometry can be accurately captured.

In short, slope surveying requires attention to both the accuracy of individual points and the density of the point cloud. While point clouds can capture countless points at once, the positional error of each point may be larger than that of conventional TS surveying. However, the large number of points provides statistical error cancellation, allowing accurate reproduction of the surface geometry. For example, fitting a cross-section to many point cloud points will average out local scatter and yield a smooth surface. Considering these characteristics, it is expected that 3D point cloud technology can achieve both cm-level accuracy (half-inch accuracy) and comprehensive shape capture for slope management.

Results and Comparison: Accuracy Differences between GNSS × Smartphone Point Clouds and Conventional Methods (TS, Drone)

We present a case where the accuracy of GNSS × smartphone point cloud measurement was actually verified in the field. On a road slope (height about 10 m (32.8 ft), length about 50 m (164.0 ft)), the entire slope was scanned with a smartphone fitted with LRTK, and the point cloud data were compared with results from total station surveying and drone photogrammetry.

First, several known points and feature points installed on the slope were selected, and coordinates read from the smartphone point cloud were compared with TS-measured values. Horizontal and vertical errors were generally within 2–3 cm (0.8–1.2 in). For example, a checkpoint at the slope crest showed a planar position difference of 2.1 cm (0.8 in) and a height difference of 2.8 cm (1.1 in), demonstrating excellent agreement with TS surveying. Regarding the entire slope shape, longitudinal and transverse sections derived from the smartphone point cloud closely matched cross-sections created by connecting TS survey points, confirming that overall geometry and even local irregularities coincided.

Next, the same slope was measured by drone photogrammetry (generating a point cloud from aerial photos) and compared with the smartphone point cloud. After aligning both point clouds to common control points and analyzing elevation differences across the slope surface, the average error was less than a few centimeters, and the maximum difference was about 5 cm (2.0 in) in areas with dense grass. This indicates that smartphone point cloud surveying captures slope geometry at a precision level comparable to drone photogrammetry. Visual overlays of the point clouds also showed the slope surface nearly coincident—without appearing as two offset layers—indicating high consistency between the two measurement results.

From these verifications, slope measurement using GNSS × smartphone point clouds can achieve accuracy comparable to conventional methods (TS and drone). Agreement with TS reference values at the level of a few centimeters and good correspondence with drone point clouds demonstrate that the method provides sufficient accuracy for as-built management applications. Moreover, smartphone measurement can be completed by a single person in a short time, offering significant efficiency benefits. The combination of this level of accuracy and operational ease is highly valuable in the field.

Procedure Reliability: Field Conditions and Measurement Procedures to Ensure Accuracy

To obtain high accuracy stably, it is important to prepare measurement conditions on site and follow appropriate procedures. First, for GNSS positioning, securing an RTK fixed solution (Fix) is a prerequisite. A triple-frequency RTK-GNSS receiver such as LRTK can track many satellites in open sky conditions and provide centimeter-level positioning as a rover. Before starting surveying, power up the equipment and confirm that correction information from network RTK (Ntrip) or the quasi-zenith satellite "Michibiki" CLAS signal is being received correctly. The fixed solution will be indicated on the smartphone screen, so always check this status. If Fix cannot be obtained, check for tall structures or trees blocking signals, change your position, or wait and retry. Even in areas outside network coverage, CLAS-compatible devices can receive augmentation signals directly from satellites, but initial convergence may take time. By remaining still and averaging for a few minutes until positioning stabilizes, reference point coordinates can be obtained with sub-centimeter horizontal accuracy and vertical accuracy on the order of a few centimeters.

There are also points to note in the smartphone point cloud capture procedure. Essentially, hold a smartphone equipped with a camera or LiDAR sensor and walk while aiming at the slope, but certain practices are required to ensure accuracy. First, scan uniformly from bottom to top so that the entire slope is covered. For long slopes, divide the area and capture in sections, providing overlapping areas to stitch the data. Move the smartphone slowly and capture features from various angles so that the point cloud generation does not create blind spots or density irregularities. The crest area is especially difficult to see from below, so if possible, circle above and scan from the crest side as well. Conversely, the toe can be overlooked if you are too far away, so approach and capture as needed. Smartphone LiDAR can miss points outdoors due to sunlight, so primarily rely on photo-based point cloud reconstruction and supplement details at close range with LiDAR.

Verification and correction of acquired point cloud data are also indispensable. After capturing a smartphone point cloud, check in software for deviations from known points. Many apps and analysis tools allow you to georeference the point cloud by adjusting it to several reference points. For example, identify existing stakes or structural reference points with known accurate coordinates in the point cloud, then translate and rotate the entire cloud so those positions coincide—this corrects small positional shifts and tilt errors. It is also effective to measure dimensions directly on the point cloud for quick on-site verification. Comparing known distances (such as spacing between toes or slope length) with point cloud measurements confirms that the scale is correctly represented. By quickly providing feedback on site and proceeding with measurement, the reliability of point cloud data can be maintained at a high level.

Implementation and Reproducibility: Repeatability and Quality Assessment in Field Cases

When introducing a new surveying method on site, it is essential to confirm reproducibility and data quality variability. Examples show that smartphone point clouds × GNSS slope measurement can produce very stable results under appropriate conditions.

One example compares point cloud models generated from scanning the same slope on two different days. Despite different operators on each day, the resulting slope point cloud geometries showed almost no differences. Distance analysis between point clouds indicated average discrepancies of 1–2 cm or less, and maximum differences were contained within about 3 cm. This demonstrates that results from this method converge to a consistent accuracy level regardless of the surveyor or timing, and that cm-level accuracy (half-inch accuracy) can be reproducibly obtained unless site conditions change significantly.

For quality assessment, it is important to check accuracy using objective indicators. For example, obtain residuals (differences) in height and position of several verification points measured by TS, and compute the RMSE (root mean square error) to understand overall accuracy trends. In measured cases, RMSE for smartphone point cloud verification points was about 2 cm on planar position and about 3 cm in height, confirming an accuracy level similar to conventional photogrammetry and mobile mapping systems. Additionally, as-built inspections confirmed that cross-sectional dimensions of slopes calculated from smartphone point clouds fell within specification values, which was independently verified by TS measurements. In other words, the data obtained by this method have sufficient reliability to withstand third-party checks.

Currently, simple 3D measurement using smartphone × GNSS is being introduced in ICT construction trial projects by the national government and local municipalities. In real projects where it has been applied to as-built measurement, results meeting inspection quality standards were achieved while greatly reducing work time compared with conventional methods. Accumulating such track records has steadily increased field engineers’ confidence in this method.

Error Factor Analysis of Measurement Results (Satellite Corrections, Vegetation, Sunlight, LiDAR Resolution, etc.)

Even with new technologies, it is necessary to understand the various error factors that affect measurement results. Below is a concise summary of the main factors, their impacts, and mitigation measures.

Satellite correction information and GNSS positioning: GNSS accuracy is heavily influenced by the availability and quality of correction information. If corrections from RTK reference stations cannot be received, or if a fixed solution is not obtained and only a float solution is available, errors of tens of centimeters can occur. Fortunately, devices like LRTK generally maintain corrected states via network RTK or satellite-distributed CLAS augmentation, but some positioning errors may still arise during periods of poor satellite geometry or ionospheric disturbances. Near slopes, one side may be blocked by terrain or structures so that only half the sky is visible; in such cases, the number of visible satellites decreases and multipath (signal reflection) becomes more likely, degrading accuracy. Countermeasures include measuring from as open a location as possible, mounting the GNSS receiver on an extended monopod to bring it closer to the sky, and averaging positioning over a longer time to reduce temporary fluctuations. Continuously monitor satellite corrections and positioning status, and exercise caution to exclude or re-measure data segments that show anomalies.

Vegetation and ground conditions: Grass or dense vegetation on a slope contributes to measurement errors. Point clouds represent the surfaces detected by lasers or photos; in grassy slopes, the data will capture blade tips rather than the bare ground, making the surface appear elevated relative to the true terrain. For example, a densely growing 10 cm (3.9 in) grass layer will make the point cloud appear as if the slope is built up by 10 cm. Ideally, mow vegetation beforehand to expose the ground. If this is impossible, avoid windy days when foliage moves, capture when vegetation is still, and filter out clearly elevated points (noise) during post-processing. Also, wet or highly reflective surfaces and very dark surfaces are hard for both laser and photogrammetry to capture accurately; avoid measuring immediately after rain or under extreme contrast between shadow and sunlight.

Sunlight and lighting conditions: Smartphone LiDAR (infrared laser) can be noisy outdoors under strong sunlight. LiDAR effectiveness and accuracy tend to degrade in bright conditions, so daytime outdoor measurements often rely primarily on photo-based point cloud reconstruction, with LiDAR used for close-range detail. Photogrammetry itself is also affected by lighting conditions. Strong midday sun can create harsh shadows and overly high contrast or cause overexposure that loses detail, while late afternoon or dusk may be too dark and increase image noise, making feature detection difficult. Ideally, measure on lightly cloudy days that provide even illumination; otherwise, shoot during morning or late afternoon oblique-light conditions. In tunnels or under bridges where sunlight does not reach, using the phone’s light or portable floodlights while capturing can prevent point cloud gaps.

LiDAR resolution and distance attenuation: Smartphone LiDAR sensors have limited output and scanning range compared with professional terrestrial laser scanners for civil engineering. Point density obtainable is not that high, and scanning targets several meters away can result in point spacing of several centimeters or more. As distance increases, the beam diverges and reflection weakens, so effective measurement range is at best about 5–10 m (16.4–32.8 ft). Therefore, attempting to scan a wide slope from a single distant location will not yield sufficient point density or accuracy. Smartphone point cloud practice requires getting as close as possible to the target and walking to perform successive close-range captures to cover the whole slope. When combining photogrammetry, the camera’s pixel resolution also affects accuracy: distant subjects appear small in images and produce coarse point clouds. Although smartphone cameras have improved, there are limits to capturing distant detail. Frame the target to fill the image at an appropriate angle and distance, and use the zoom function if necessary. The key point is that, given equipment-specific resolution and range constraints, the closer and more detailed the capture, the better the accuracy.

By understanding and addressing these factors, the reliability of smartphone × GNSS slope measurement is further enhanced. Modern software increasingly includes AI-based noise removal and positional correction functions, and with appropriate post-processing, residual errors can be reduced to within a few centimeters. On site, keeping these error factors in mind and performing additional measurements or calibrations as needed will ensure the final deliverables meet accuracy requirements.

Summary: Rationality and Value of Introducing Smartphone × GNSS

As described above, combining smartphone point cloud capture and GNSS positioning can provide a rational and useful solution for slope surveying. The rationale mainly condenses into the following points:

• Balance of accuracy and coverage: It is groundbreaking in that it achieves cm-level accuracy (half-inch accuracy) while remaining easy to use and enabling a single person to cover wide areas. By recording the entire slope as a high-density point cloud while ensuring sufficient accuracy, it contributes to both quality management and efficiency.

• Improved operational efficiency and safety: With just a smartphone and small GNSS device, there is no need to carry heavy equipment up the slope. Work that previously required multiple people can be completed by one person in a short time. Collecting data remotely without high-elevation work greatly reduces risk and facilitates frequent progress checks and as-built verifications.

• Cost-effectiveness and low barrier to adoption: Initial investment is lower than procuring dedicated 3D scanners or expensive surveying instruments, and existing smartphones can be used, simplifying equipment management. Apps are intuitive and training costs are small; internalizing 3D measurement reduces outsourcing costs and enables rapid decision-making through immediate data sharing.

Accordingly, smartphone × GNSS slope measurement strikes a high-level balance of high accuracy, safety, and low cost, and can strongly support DX (digital transformation) on construction sites. Making 3D measurement—previously outsourced to specialist companies—easily executable in-house can reduce costs and improve responsiveness, marking the first step toward advanced, data-driven construction management.

Conclusion: Toward the Next Phase of Slope Management with Simple LRTK Surveying

Slope surveying and management have long required time and expertise. However, the approach of smartphone surveying using LRTK is bringing an era in which anyone can easily perform precise 3D measurement. Slope point cloud data obtained at cm-level accuracy (half-inch accuracy) can be directly used for design comparisons, volume calculations, and deformation monitoring. For example, regular scans of slopes after construction can detect minor landslides or deformations in digital data and inform early repair decisions. This represents a new management method that was not achievable by conventional visual inspection or partial surveying.

The spread of simple GNSS × point cloud surveying will elevate slope management to the next phase. Field personnel themselves will be able to immediately acquire and share high-accuracy terrain data, enabling consistent, data-driven decision-making from construction control through maintenance. Slope management that once relied on intuition and experience will evolve into an evidence-based scientific process. By wisely incorporating advanced technologies such as LRTK and accelerating the DX of slope management, we can take a step toward next-generation safe and reliable infrastructure maintenance and management.