7 Practical Procedures for Drone Point Cloud Analysis to Detect Slope Deformations

Table of Contents

• Step 1: On-site investigation and development of the surveying plan

• Step 2: Drone flight planning and preflight preparation

• Step 3: Installation of control points (GCP) and coordinate measurement

• Step 4: Acquisition of slope surface data by drone

• Step 5: Generation and preprocessing of point cloud data

• Step 6: Analysis and detection of deformations by point-cloud differencing

• Step 7: Preparation of the results report and application for maintenance management

First, in slope management (norimen kanri), using 3D point cloud data acquired by drone-mounted cameras and LiDAR makes it possible to detect fine changes that conventional methods might overlook. The Ministry of Land, Infrastructure, Transport and Tourism's i-Construction initiative encourages the use of UAVs (drones) in public surveying, and the Geospatial Information Authority of Japan has prepared a manual specifying procedures and accuracy control for drone photogrammetry.

Point cloud surveying can densely measure slope surfaces with countless points, allowing local irregularities and displacements to be recorded without omission. For example, the Ministry’s "3D As-built Management Guidelines (draft)" indicates a standard for measuring slope surveys using terrestrial laser scanners at a point cloud density of 100 points or more per 1 m² (10.8 ft²). Covering the entire slope with points without gaps aims to capture even small bumps and hollows accurately, and similarly, acquiring high-density data is important for drone point cloud monitoring.

This article introduces seven concrete procedures for condition monitoring using drone point cloud analysis, aimed at practitioners involved in slope maintenance work. Step by step, it explains points for safely and accurately digitally recording the current condition of slopes and for not missing signs of long-term changes.

Step 1: On-site Survey and Development of the Surveying Plan

First, begin by assessing the site conditions and formulating a plan. Investigate in advance the location and scale of the slope and the surrounding environment, and confirm whether drone flights are feasible and where safe takeoff and landing areas are. If necessary, conduct a site reconnaissance to understand the extent of tree growth and geomorphological risks such as scree or colluvium. At the same time, clarify the purpose of the survey (e.g., as-built verification after construction or displacement monitoring through regular inspections) and determine the accuracy requirements and scope of the data to be acquired. For example, if the aim is monitoring long-term changes, acquire an initial baseline point cloud dataset so that it can be compared with subsequent periodic measurement data.

Selecting the measurement method to be used is also an important step. Generally, photogrammetry using drone-mounted cameras (photogrammetry) is widely used as a low-cost method to acquire extensive 3D point clouds without expensive laser equipment. In photogrammetry, to obtain sufficient accuracy it is necessary to keep a high overlap rate of the captured images (described below), but recent advances in software make it possible to achieve practical accuracy with easy operation. On the other hand, when the target slope is covered by forest and capturing the ground surface is difficult, or when a higher-density, higher-accuracy point cloud is required, mounting LiDAR (laser scanner) on a drone is also considered. Laser surveying can be performed at night and in light rain, and because some of the laser reaches the ground through gaps in the trees, it has the advantage of obtaining accurate ground-surface data even on slopes covered with forest or grass. However, LiDAR equipment is expensive and requires specialized skills for operation and analysis, so many companies currently outsource to specialist contractors rather than owning the equipment themselves. Based on site conditions, budget, and personnel skill levels, choose the optimal measurement method and decide on the equipment to be used (drone model, camera, or LiDAR) and any necessary ancillary equipment. Finally, prepare a surveying plan that reflects these decisions and organize the flight area and altitude, shooting method, required personnel, schedule, and so on. For public projects, share the plan with the client in advance and, if necessary, apply for flight permissions and notify relevant parties.

Step 2: Drone flight planning and pre-flight preparations

To obtain data safely and efficiently, prepare thoroughly by creating a concrete flight plan. Following the survey plan drafted in Step 1, finalize the drone’s flight paths, altitudes, and imaging settings in detail. To cover the entire slope without omission, it is important to set appropriate flight routes and shooting angles according to the shape of the slope. Vertical overhead nadir images alone do not capture the sides of steep slopes sufficiently, so combine them with shots taken with the camera angled somewhat obliquely (i.e., facing the slope), and plan flight paths that acquire images from multiple directions. For example, when flying near the toe of the slope, tilt the camera toward the slope to capture the side, and for the mid- to upper-slope areas shoot looking down from above; design a route that varies altitude and camera tilt angle to comprehensively cover the whole slope.

Next, prepare to carry out the planned flight safely. First, inspect and service the aircraft and batteries. Check the drone’s operation, clean and verify the installation of propellers and sensors, and thoroughly confirm battery charge levels and degradation. If necessary, prepare spare batteries and ensure you have the capacity to photograph the entire area as planned within the flight time. Also perform pre-flight checks of GPS reception and magnetic compass calibration to ensure stability during flight. In addition, check the weather conditions in the flight area. Because strong winds or rain can degrade image quality or cause aircraft accidents, monitor the day’s weather forecast closely and remain flexible to set a backup date if needed. Before flight, implement safety measures such as restricting access to the area and third-party monitoring, and share emergency procedures (responses to aircraft troubles, communication protocols, etc.) with all stakeholders. Finally, verify the firmware and flight program settings of the drone to be used, and conduct test flights in advance (in a similar environment if possible) to confirm that photographs can be obtained according to the plan. Such thorough preparation prevents unexpected troubles during the actual flight and enables data acquisition as planned.

Step 3: Placement of Ground Control Points (GCPs) and Coordinate Measurement

Securing ground control points (Ground Control Point, GCP) is indispensable for obtaining high-precision point cloud models. A ground control point is a reference point whose accurate three-dimensional coordinate values (planar position and elevation) are known in advance, and it serves to provide point cloud data generated by photogrammetry with an accurate position in absolute coordinates (such as public coordinate systems). First, select locations suitable for GCPs within the survey area and install conspicuous aerial targets (target markers) there. GCPs should be placed so as to surround the survey area as much as possible, and when slope geometry has undulations, it is desirable to distribute them at the upper and lower edges and at locations with large irregularities. In fact, the Geospatial Information Authority of Japan’s Public Surveying Manual (UAV Photogrammetry) also prescribes that “GCPs should be placed in positions covering the entire survey area while considering the shape of the measurement target area and changes in relative elevation.” Measure the installed GCPs’ XYZ coordinates using a high-precision GNSS receiver (RTK positioning) or a total station so they can be used as reference points in later analyses.

In addition to control points, validation points (checkpoints) for verifying analytical accuracy are also established as needed. Validation points are separate from control points and are independent points used to check the model’s accuracy; by comparing coordinates derived from the generated point cloud with the measured values at the validation points, errors can be assessed. The number of control and validation points depends on the survey area and required accuracy, but it is generally recommended to place at least 5 points and allocate more than half of them as validation points. For example, if 7 points are installed, 4 points should be control points and 3 points validation points, distributed evenly across the entire area. This allows balanced verification of accuracy across the whole model. In recent years, aircraft equipped with RTK-GNSS that can record camera coordinates for each photo with high precision during flight, and PPK technologies that perform high-precision trajectory correction in post-processing, have emerged; using these can in some cases greatly reduce the number of control points required. There is also movement within the Ministry of Land, Infrastructure, Transport and Tourism (MLIT) guidelines to allow omitting the placement of control points when using direct positioning methods such as RTK-equipped drones. However, it is desirable not to reduce them to zero and to leave at least 1–2 points for validation; if no control points are placed at all, care must be taken to ensure there are no systematic errors across the entire model. As described above, appropriate placement of control and validation points and precise positioning to obtain coordinates lay the groundwork for constructing high-accuracy point cloud models in subsequent photo analysis.

Step 4: Collecting slope data using a drone

It is finally time to fly the drone and acquire image data or laser data of the slope. Take off after completing safety checks in accordance with the pre-established flight plan. During flight, it is efficient to stabilize the aircraft in GPS mode and use a waypoint navigation program for automated flight. The operator constantly monitors the aircraft’s altitude and position and verifies that the planned survey area is being covered. If unforeseen situations occur—such as the aircraft being blown off course by unexpected strong winds or birds approaching—do not force the operation; pause temporarily and reestablish a safe operating posture.

For photogrammetry, it is essential to acquire a sufficient number of photographs to capture the entire slope. Photographs should be taken so that adjacent images have at least 80% overlap, and preferably 85% overlap . A high overlap rate allows many matching points to be detected during image analysis, improving the density and accuracy of the point cloud. Also, ensure lateral overlap between flight lines, and for areas that are difficult to photograph, such as the upper and lower edges of the slope, perform additional shooting from different angles and distances. For example, reduce blind spots by adding oblique shots from above for the slope crest (upper edge of the slope) and shots looking up from below for the slope toe (lower edge of the slope). Regarding illumination, a bright time of day is desirable, but because shading on a slope can differ between morning and afternoon, choose a time when lighting conditions are as uniform as possible. If possible, set the shooting mode to manual and fix the exposure so that brightness does not vary significantly across all photos. After finishing photographing the entire slope, check the number of photos and the coverage to ensure nothing is missing.

In laser surveying (UAV LiDAR), the aircraft is autonomously flown along the planned flight route while a LiDAR sensor acquires laser point clouds. Because laser measurements continuously capture large amounts of point cloud data during flight, operators should pay attention not to miss the measurement on/off timing. For example, start laser measurements a few dozen meters (a few dozen ft) before the survey section and stop with a margin beyond the endpoint to ensure the target area is fully covered. For drone LiDAR, also set overlapping areas across multiple flight lines and illuminate from different angles to fill in parts that would be blind spots from one direction, such as beneath bushes and behind rocks. During acquisition, if a real-time point cloud preview function is available, use it to confirm that points are being recorded properly. Unlike photogrammetry, measurements are generally fine under cloudy conditions, but strong rain should halt operations because lasers can scatter and cause noise. While coordinating with assistants as required for safety and legal reasons, land the aircraft once data acquisition for the planned area is complete. Finally, verify that captured images and measurement data have been saved and back them up as needed. At this point, the raw data needed for analysis has been collected.

Step 5: Point Cloud Data Generation and Preprocessing

From acquired image data or laser survey data, we generate 3D point clouds and prepare them into data suitable for analysis. For photogrammetry, the captured photos and the control point coordinates are input into specialized SfM/MVS software, which performs point cloud generation by photo analysis. The software matches feature points between photos to estimate camera positions and orientations, and from that reconstructs 3D point clouds on the order of several million points. Recent software often automates the workflow so that, once you load the photos and provide known point information, the entire process (feature point extraction → alignment → dense point cloud generation) can proceed with a single click. Analysis time depends on the number of photos and PC specs, but typically takes on the order of tens of minutes to several hours.

After processing, we perform georeference (position alignment) of the generated point cloud data using the coordinates of pre-measured control points (GCPs). This assigns absolute coordinates to the point cloud, allowing the slope point cloud model to be correctly positioned in any desired coordinate system (such as the Plane Rectangular Coordinate System ◯ system). We also match the coordinates of check points and calculate and verify the errors on the point cloud model. If the errors fall within the allowable range (for example, within a few cm (within a few in) horizontally and vertically, respectively), the photo analysis is successful. If they exceed the limits, we consider reprocessing, such as adding more photos, revising feature point settings, or excluding photos with extreme errors.

For laser survey data, the trajectory data from the GNSS/IMU mounted on the drone and the laser range measurement data are integrated to create the initial point cloud. Next, post-processing corrections to reference points (PPK) are applied, and if multiple flights were conducted the point clouds are merged and aligned (registration) to integrate them into a single point cloud dataset.

After that, unnecessary points and noise not representing the ground surface are removed. For example, point clouds capturing structures around the slope, heavy machinery, passing vehicles, or people are deleted because they hinder analysis. Laser point clouds may also include noise from birds or raindrops in the air, so clearly isolated points and outliers are removed by filtering. Similarly, photogrammetric point clouds are trimmed to remove points outside the target, such as sky or background buildings. Additionally, coordinate transformations and adjustments to the vertical datum may be performed based on surveying standards. For public surveying, transformations between geodetic datums (World Geodetic System ⇔ local coordinates) and reference height corrections for creating longitudinal profiles may be required.

Finally, the cleaned point cloud data are exported in formats appropriate to the intended use. Common formats are LAS, PLY, and XYZ, and colored orthophotos or 3D mesh models may also be exported for visualization. This is the point cloud generation and preprocessing workflow. At this stage, a high-precision three-dimensional point cloud model faithfully reproducing the entire slope has been obtained. Thereafter, this point cloud data is used to compare with past data and perform various measurements.

Step 6: Analysis and Detection of Deformations by Point Cloud Differencing

Analyze the acquired point cloud data to extract slope deformations (changes and anomalies) without missing them. To avoid “missing deformations,” comparing point cloud data in a time series is effective. Use the point cloud from the initial survey as the reference model, and by comparing it with point cloud data acquired again after a certain period using the same procedure, you can quantitatively grasp the amount of change in the slope geometry. Specifically, overlay the old and new point clouds from the two periods (adjusting coordinates), and calculate the differences in height and position of corresponding parts. In general, anomaly detection using point cloud data commonly uses differential analysis to extract deformations.

For example, if you convert the point clouds to a grid-shaped mesh surface and then compute the vertical distance differences, you can visualize, as areas, how much the terrain has uplifted or subsided at a given location.

Even for a vast collection of points covering an entire slope, taking differences can highlight the locations where displacement has occurred.

In practice, differential analysis is performed using commercial point cloud processing software or GIS software. For example, the reference-time point cloud is converted to a TIN (triangulated irregular network) or a gridded surface, and the distance from each point in the later-time point cloud to the reference surface is calculated (positive indicates erosion, negative indicates accumulation). Mapping the results with colors makes it possible to intuitively grasp the height changes of the slope surface. For example, in a colored heat map, stable areas with change amounts near 0 are shown in green, receding (erosion) areas are blue to purple, and advancing (fill or collapse deposition) areas are red to yellow, allowing immediate identification.

From the difference map, small changes that were previously easy to overlook can also be detected, such as surface delamination, sediment accumulation, and localized subsidence caused by new cracks. Also, using the volume calculation function, the amount of collapsed soil can be computed in an instant.

For example, because it can quantitatively show even that "○○ cubic meters (○○ ft³) of soil were lost from the slope shoulder due to the collapse," it directly links to the planning of restoration works.

Monitoring by successive scans, not just differential analysis, is also effective for detecting deformations. If point clouds are acquired and overlaid at each regular inspection, you can accumulate a history of which positions within the slope have moved and by how much. Areas with particularly notable displacement tend to show some kind of precursor. For example, if data capture reveals changes such as "a central area of the slope is continuously retreating by several centimeters (several in)" or "a slight uplift is observed around specific retaining blocks," early consideration of reinforcement work or detailed investigation becomes possible.

Thus, deformation analysis using drone-derived point clouds provides objective and quantitative evidence compared with traditional inspections that rely on human visual observation. Of course, to ensure the accuracy of the analysis results, the quality of the input data is a prerequisite, so it is important to reliably carry out the accuracy control described in the previous procedures (calibration, registration/alignment, noise removal, etc.). Also, locations where changes are observed in the differences must be cross-checked against actual photographic images and on-site conditions; the process of having technicians verify whether they are real deformations rather than false changes (survey errors or temporary phenomena) is indispensable. Ultimately, the suggestions obtained from point cloud analysis and the findings from field inspections are integrated to make a judgment and lead to the planning of necessary countermeasures.

Step 7: Preparing the Results Report and Applying It to Operation and Maintenance

We convey the slope conditions discovered by analysis to stakeholders in an easy-to-understand manner to support future maintenance and management. Measurement results and information on locations of deterioration obtained from point clouds are compiled into drawings and reports. For example, three-dimensional models and orthophotos that show the current condition of the slope are useful as-is for reporting materials. Longitudinal and cross-sectional views can be created from the point cloud at arbitrary positions, and by overlaying them with the design cross-section lines, construction errors and deformation amounts can be presented as concrete numerical values.

In addition, we paste a difference heat map from the previous survey and add comments such as “a maximum subsidence of ◯ cm (◯ in) was confirmed near ○○” and “retreat due to surface soil washout at part of the slope shoulder.” For as-built management using point cloud measurements, as-built drawings and quantity calculation sheets are also prepared as needed. For example, for slope protection works (mortar spraying, etc.), the surface area can be calculated from the point cloud to determine required material quantities, and for embankment slopes, the as-built thickness can be evaluated.

Furthermore, if there are collapsed sections, the volume of soil can be calculated to serve as materials for considering selection of restoration methods.

When preparing a report, the key point is to convey information visually by effectively using diagrams and photographs rather than merely listing numbers. Including images such as aerial drone photographs marked to show areas of deterioration and capture images of 3D models generated from point clouds makes it easier for senior management and third parties who are unfamiliar with the site to understand the situation. If necessary, consider delivering or sharing the actual point cloud data and orthoimages as deliverables. In recent years there have been services that allow sharing and viewing point clouds and 360° images on the cloud, enabling recipients to check 3D data via a browser even without a high-performance PC.

By leveraging these systems, managers and specialists can check detailed site data from the office, which speeds up decision-making and facilitates consensus-building.

Finally, incorporate the report's findings and recommendations into future maintenance plans. For example, consult with relevant departments on countermeasures such as "conduct ongoing monitoring of the slope every six months" and "initiate repair design promptly for areas showing significant deterioration." The objective data obtained from drone point cloud analysis also serve as supporting evidence to justify repair works. By linking this series of results to subsequent actions in this way, the value of drone-based slope inspections is maximized.

This article has explained the practical procedures for slope point-cloud analysis using drones in seven steps. Even minute changes that were previously overlooked by manual and cross-sectional surveys can be clearly revealed by differencing high-density data when drones and point-cloud technology are combined. With proper planning, accuracy control, and safe operation, this becomes a powerful means to assess slope condition with objective data and support preventive maintenance. Please refer to the procedures in this article when considering the use of drone point clouds on site. As increasingly accessible and higher-performance measurement technologies emerge, the DX (digital transformation) of slope management operations is expected to accelerate further, contributing to the safe and efficient maintenance of infrastructure.