Five Ways to Improve Accuracy in Slope Inspections: Streamline with Drone Point Cloud Analysis

Table of Contents

• Make meticulous advance plans

• Ensure high-quality data acquisition

• Use ground control points and high-precision positioning

• Minimize errors in point cloud data processing

• Validate analysis results and pursue continuous improvement

Summary

A slope (norimen) refers to an artificial incline found along roadsides and on developed land, and it plays an important role in preventing landslides. Regular inspections are essential to confirm that slopes have been constructed to the design shape and that no deformation or collapse has occurred due to aging or heavy rainfall. The objectives of slope inspections are broadly divided into two. One is as-built verification during construction: measuring whether slopes formed by embankment or cutting match the gradients and heights shown on the design drawings to confirm construction quality. The other is maintenance after opening: periodically inspecting completed slopes for deterioration or deformation due to aging, and planning repairs when necessary. When slopes collapse due to earthquakes or heavy rain, it is also important to quickly determine the volume of collapsed soil and the affected area to enable safe restoration planning. Traditional slope inspections and surveys required workers to enter the slope and measure a few points with a total station, or to measure slope angles with tape measures and inclinometers. However, this approach has limits in capturing the entire slope in detail, requires a large amount of manpower and time, and work on steep slopes is always hazardous.

Recently, a method has gained attention that acquires and analyzes the entire slope as point cloud data using photogrammetry or laser scanning from drones (unmanned aerial vehicles). By analyzing high-density 3D point cloud data obtained from drone aerial photography, you can capture the overall shape of a slope safely and in a short time, enabling precise inspections without misses. In practice, in some sites slope surveys that used to take half a day have been completed in several tens of minutes with drones, and safety has greatly improved because people do not have to enter the slope. Moreover, aerial surveys can be performed by a single person, reducing staffing burdens and allowing freed personnel to be assigned to other tasks. Drone point cloud analysis can thus realize more efficient and advanced slope inspections, but certain measures are necessary to maximize accuracy.

To read accurate dimensions and displacements from point cloud data acquired by drones, it is important to follow appropriate procedures at each stage of surveying and analysis and eliminate error sources. If procedures are mishandled, distortions or positional shifts can occur in the point cloud, undermining accuracy and rendering the data unreliable. This article therefore explains in detail five key points for improving the accuracy of slope point cloud analysis. By mastering tips for each process—from advance preparation through surveying, data processing, and result validation—you can achieve efficient yet highly accurate slope inspections.

• Make meticulous advance plans

The first step to obtaining high-accuracy point cloud data is careful planning before surveying. Before going to the site, understand the slope’s shape, extent, and surrounding conditions, and consider thoroughly how to fly the drone. Plan flight routes that will cover the entire slope without gaps so there are no blind spots or missing data in the capture area. Especially for steep, tall slopes, photographing only from directly above may fail to capture the slope’s side faces sufficiently. In such cases, include oblique photography in the flight plan by tilting the drone camera so the slope’s berms and faces are captured in detail. Capturing from multiple angles and paths improves 3D reconstruction accuracy and yields point clouds that enable precise determination of slope geometry in later analyses.

Flight altitude and image overlap rate also greatly affect accuracy. In general, photogrammetry prefers image overlap of 80% or more between adjacent photos, and for wide areas plan flight courses to overlap in a zigzag pattern. Set flight altitude according to the desired ground sampling distance (GSD); if you want to detect small displacements or cracks, fly as low as possible. For large slopes, divide the work into multiple flights and ensure overlapping regions so point clouds can be merged later. Also plan for on-site safety management and flight permissions—determine vehicle staging areas and control third-party access to the site. Thorough advance planning enables efficient surveying and minimizes the risk of data shortages or having to re-fly, thereby reducing error risk.

At the planning stage, selecting the surveying method and equipment is also important. For example, if the slope is covered with trees, photogrammetry alone may not yield sufficient ground surface points. In such cases consider equipping the drone with LiDAR (laser scanner). Conversely, on bare rock slopes with few surface patterns or features, photogrammetry may struggle with feature matching and thus suffer reduced accuracy; in such cases placing visible targets on the slope can help. By choosing the best method and equipment for site conditions, you can efficiently acquire high-quality point cloud data.

• Ensure high-quality data acquisition

Once you have a detailed plan, strictly manage quality during actual drone flights. Before starting flights, calibrate the aircraft and sensors, and confirm that GPS and IMU are functioning properly. For camera use, check lenses for dirt and, if necessary, set focus and exposure manually so you obtain clear images. Use sufficiently fast shutter speeds and keep drone speed appropriately low to prevent blur and distortion. Rolling-shutter cameras are particularly susceptible to image distortion caused by high-speed flight, which can reduce accuracy; in such cases reduce speed or, if possible, use equipment with a global shutter. To avoid optical noise, take care to avoid shooting into strong direct sunlight that causes extreme contrast or lens flare.

During flight, adhere precisely to planned routes and altitudes, and monitor to ensure required overlaps are achieved. Slope airspace is often wind-prone, so avoid flying in strong winds and choose conditions that allow stable sensor measurements. Rain, fog, or extreme contrast conditions can also disturb sensor accuracy, so pay attention to weather. Moving objects within the measurement area—construction vehicles, heavy machinery, or pedestrians—can introduce noise into point cloud data. Ensure safety and, ideally, select times when people and vehicles will not be present in the capture area.

Sharp images are crucial for photogrammetry, so avoid mixing motion-blurred or out-of-focus photos; if in doubt, retake images on-site. For laser scanners, perform sensor initialization and test scans before flight to confirm accurate point measurement. When capturing point clouds over multiple flights, ensure sufficient overlap between flights so merging later does not produce misalignment. The higher the quality of raw data, the less you will have to struggle with correction in later steps, and the more accurate the point cloud model you can build. Striving for high quality during on-site data acquisition is the fastest route to improved accuracy.

• Use ground control points and high-precision positioning

To handle point clouds from a drone with correct dimensions and coordinates, using ground control points (GCPs) as survey references is indispensable. Creating a 3D model using only the drone’s onboard GPS can result in positional shifts of several meters (several ft) or slight scale errors across the model. To correct this, place several markers at known coordinates on site and align the point cloud to those coordinates. For example, place about 3–5 GCPs around the slope’s perimeter and at the top and bottom, and measure them with high-precision GNSS surveying or a total station. Use markers (targets) that are easily identifiable in aerial photos and ensure sufficient satellite reception time to obtain centimeter-level coordinate values (half-inch accuracy). Input these reference points into point cloud processing software to georeference the aerial photogrammetry, which aligns the model accurately to a public coordinate system or a chosen local coordinate system and yields precise 3D dimensional data. Georeferenced point clouds can be easily overlaid with design drawings or other terrain survey data, making them useful for as-built inspections and monitoring for changes over time. Distribute GCPs evenly around the slope to prevent model distortion—on wide slopes place them at upper, lower, and central areas.

Recently, drones equipped with RTK-GNSS for high-precision positioning during flight have become widespread, allowing substantial reduction in required GCPs. Using RTK-capable drones or PPK post-processing for precise trajectory correction greatly improves the positional accuracy of aerial images. As a result, fewer reference points can still yield high model accuracy, reducing the need to place many markers on unstable slopes. However, even when using RTK-equipped drones, it is advisable to set a few reference points or check points for verification to confirm the finished point cloud’s accuracy. As a new technology to reduce the work of GCP surveying, small high-precision GNSS receivers that attach to smartphones have also appeared. Smartphone-based RTK positioning enables a single person to easily measure centimeter-level reference points (half-inch accuracy), dramatically streamlining coordinate verification tasks required for slope inspections. Proper use of ground control points and ensuring positioning accuracy are key to improving point cloud analysis accuracy.

• Minimize errors in point cloud data processing

There are many factors affecting accuracy during the stage of building and analyzing point cloud models from on-site data. When modeling with photogrammetry or point cloud processing software, first set the coordinate and unit systems correctly and ensure GCP information is properly applied. In photogrammetric alignment (aerial triangulation), check the camera intrinsic parameters and lens distortion corrections calculated by the software for unexpectedly large distortions or variances. If needed, input additional GCPs or use high-precision computation options to remove global distortion or scale errors from the model. When converting GNSS-derived heights to a public survey vertical datum, remember to apply geoid model transformations. When merging point clouds from multiple flights or different equipment, perform registration rigorously in overlapping regions. If initial position offsets are large, start with a coarse alignment using a low-density point cloud and then apply precise alignment (e.g., the ICP algorithm) on high-density data to ensure correct integration. After merging, check for unnatural steps or duplicated surfaces and perform further positional corrections if necessary.

Post-processing such as noise removal and classification is also important for the resulting point cloud. Slopes are often surrounded by grass and shrubs; leaving vegetation points in the point cloud makes it difficult to determine the true ground surface. Use automatic classification functions or filters in dedicated software to remove vegetation points and extract a ground point cloud (digital terrain model), which makes it easier to quantitatively evaluate slope gradient and deformation. However, be careful not to over-filter; removing necessary points can erase fine details of the shape and degrade accuracy. Carefully remove only noise and outliers while preserving slope edges and important structural points to maintain a balance. Thorough quality control during processing maximizes the potential of the raw data and produces a reliable point cloud model.

In photogrammetry, when the capture area is vast and GCPs are insufficient, the model can warp in a phenomenon called the "bowl effect." To prevent such distortion, sufficient photo overlap and proper GCP placement for correction are critical. In laser scanning, slight offsets in the aircraft’s attitude sensors or time synchronization can produce steps in the point cloud. Regular equipment calibration and ensuring overlap during data acquisition help minimize such systematic errors.

• Validate analysis results and pursue continuous improvement

Always validate and confirm the reliability of results obtained from point cloud analysis. For example, check whether slope gradients or volumes calculated from the point cloud deviate significantly from traditional survey values or design values. If you placed verification points (checkpoints) beforehand, compare their coordinate values to the corresponding points on the point cloud to confirm errors are within acceptable limits. Record mean and maximum errors across multiple checkpoints to quantify survey achievement accuracy. If errors larger than expected are found, review the causes: Was there error in GCP surveying? Did insufficient photo overlap cause local distortion? Was there a coordinate system setting mistake during software processing? Re-examine each step and consider reprocessing data or conducting supplemental surveys as needed. Such accuracy verification is emphasized in public 3D surveying guidelines, and data without appropriate verification cannot be considered reliable. Also, include accuracy information obtained through validation in reporting to demonstrate the quality of surveying results to clients and stakeholders, thereby enhancing trust in 3D point cloud data.

Validation results provide valuable feedback for future projects. Analyze optimal methods for different site conditions and equipment and accumulate lessons learned. For example: "For a slope of this size, this many GCPs were sufficient"; "LiDAR was more efficient because the slope was covered with vegetation"; "This site tends to be backlit in the afternoon, so schedule shooting in the morning." Applying such site-specific learnings improves accuracy and efficiency over time. Establish checklists and manuals within the team to share experience. Keep abreast of technological trends and be willing to evaluate and adopt higher-performance equipment and analysis software when they emerge. Through steady validation and improvement, drone point cloud analysis for slope inspections will become increasingly reliable and a staple tool on site. When frequently scanning the same slope to monitor changes over time, be sure to acquire and integrate data in the same coordinate system each time. With solid GCP and coordinate management, time-series comparison of point clouds can detect slight displacements of a few centimeters (a few in), aiding maintenance.

Summary

Introducing drone point cloud analysis into slope inspections enables you to assess slope conditions with unprecedented speed and safety. Public works agencies, including the Ministry of Land, Infrastructure, Transport and Tourism, are promoting as-built and maintenance management using 3D point clouds, and drone surveys are required to meet precision control comparable to conventional surveys. By practicing the five methods described here, you can maximize the benefits while ensuring the reliability of measured data. With careful planning, high-quality data acquisition, precision correction using GCPs, and careful data processing and validation, slopes can be measured with accuracy comparable to traditional surveying even when using drones. In fact, there are increasing cases where these methods have achieved as-built measurements within a few centimeters (a few in) in both planimetric and elevational accuracy, and drone point cloud surveys are reaching a level comparable to conventional methods. With accurate digital data, comparison with designs and detection of changes over time become easy, dramatically improving the efficiency and quality of inspection work.

Recently, new tools that support surveying operations have also emerged. For example, using a high-precision GNSS receiver attachable to an iPhone such as "LRTK" makes RTK positioning—previously requiring specialized equipment—easily achievable, allowing one person to rapidly perform GCP surveys and on-site coordinate verification. Centimeter-level positioning, once the domain of experienced technicians with expensive equipment, is now achievable on small sites and by small teams thanks to such devices. Combining these innovative devices with drone point cloud analysis further streamlines slope inspection workflows and enables decision-making based on highly accurate data. Adopt precision-improving measures for drone point cloud surveying and the latest technologies to realize safe, rapid, and precise slope inspections. Drone and ICT-enabled 3D measurement are becoming a new standard in slope management. Master inspection methods that balance accuracy and efficiency and promote on-site digital transformation (DX). Through accurate and efficient inspections, we can firmly support infrastructure safety and improve on-site work efficiency.