The SfM Processing Revolution Supported by LRTK: Achieving Centimeter-Level Accuracy with GNSS Integration

SfM processing (short for Structure from Motion) is a photogrammetric technique that reconstructs the three-dimensional shape of an object from multiple photographic images. It identifies common feature points across photos taken from different viewpoints and, based on those correspondences, simultaneously computes camera positions and orientations as well as the 3D coordinates of the feature points (bundle adjustment). This computation automatically generates a high-density point cloud (a collection of many 3D measurement points) from the photos. Thanks to advances in SfM processing algorithms, 3D modeling that once required specialized software and manual work can now be done easily, leading to widespread use of SfM-based point cloud generation in civil surveying, construction site as-built management, and other fields.

Point clouds obtained by photogrammetry can record ground topography and structure shapes in detail, with accuracies reaching millimeter- to centimeter-levels. Applications of SfM processing continue to grow each year, from creating terrain models from drone aerial photos to measuring structural dimensions from ground-based photography. A major advantage over specialized LiDAR equipment is that SfM enables low-cost and flexible 3D measurement. However, point cloud models generated by SfM initially exist in an arbitrary-scale relative coordinate system, so georeferencing is required to use them as actual survey results. The next section explains this accuracy challenge and the traditional method of installing GCPs (ground control points).

SfM accuracy challenges and the burden of installing GCPs (ground control points)

To use an SfM-derived point cloud model as survey output in a mapped coordinate system (for example, a terrain map with elevation or an as-built plan), the model must be given the correct scale and absolute coordinates. Traditionally, this is achieved by placing multiple known coordinate points on site called “GCPs (Ground Control Points)” and using them to align the model’s overall position and scale. For example, when performing drone photogrammetry over a wide development site, targets (GCP markers) are set out every several tens to hundreds of meters, and their accurate coordinates are measured in advance using GNSS surveying or a total station. In the SfM processing software, the locations of these GCPs as they appear in the photos are marked, and during analysis the model is matched to those reference points to reproduce the terrain model with centimeter-level accuracy.

However, the task of installing GCPs entails substantial effort. Placing control points across a large area requires manpower and time, and in steep or hazardous areas setting and surveying targets can be difficult. As construction progresses and terrain changes, installed GCPs may be displaced or lost, requiring re-surveying. The process of finding and marking targets in images is also cumbersome, and accuracy can be affected by operator skill and human error. Against this background, the key challenge on site has been how to achieve high-precision SfM point clouds with as few GCPs as possible.

Complementarity of GNSS positioning (RTK/PPK) and SfM: the importance of coordinate correction and scale assignment

A promising approach to address the above challenges is direct georeferencing through the combination of GNSS positioning and SfM. Using high-precision GNSS techniques such as RTK (Real-Time Kinematic) or PPK (Post Processed Kinematic), camera positions at the time of photography can be measured with centimeter-level accuracy and recorded as geotags for each photo. In SfM analysis, camera positions and orientations (external parameters) are normally treated as unknowns and solved simultaneously, but providing highly accurate initial coordinates reduces the degrees of freedom in the computation and resolves the model’s scale ambiguity. In other words, by supplying the photo data with the correct answers for scale and position in advance, high accuracy can be achieved with few GCPs, and in some cases GCP-less (no GCP) workflows can produce practically usable 3D survey outputs.

RTK-GNSS-equipped drones (UAVs) are already available, enabling high-precision coordinates to be appended to photos in real time during flight. For ground photography, attempts have been made to record shooting positions with GNSS receivers integrated into cameras. In this way, GNSS and SfM complement each other’s weaknesses. SfM alone can produce detailed point clouds but requires effort to determine absolute coordinates and scale; GNSS alone may suffer in shaded or indoor environments but provides high-precision positions in open-sky conditions. Combining both enables flexible photogrammetric capture along with reliable positioning accuracy, dramatically expanding the applicability and trustworthiness of photogrammetry.

How LRTK adds high-precision geotags and simplifies GCPs

A solution that makes this GNSS×SfM fusion easy to deploy on site is LRTK. LRTK (the name of a positioning device/service offered by Refixia Co.) consists of a compact RTK-GNSS receiver that can be attached to a smartphone and a dedicated app; it performs centimeter-accuracy positioning in real time during photo capture and records high-precision coordinates (geotags) for each photo. For example, a product called LRTK Phone allows a thin RTK receiver to be attached to an iPhone or iPad with one touch; by aiming the phone camera and pressing the shutter, latitude, longitude, and height at the shooting location are instantly obtained. The pocket-sized device weighs about 125 g, making it an easy-to-carry means to take measurements on site whenever needed.

Because LRTK can provide precise coordinates for each photo, it can greatly simplify the need for GCP installation in photogrammetry. Where a dozen or more reference points were once required, only a few verification points may now suffice, and in some cases GCPs can be completely unnecessary. The geotag information provided by LRTK is used in bundle adjustment during SfM processing to fix or serve as initial values for each photo’s position, and the point cloud model is automatically reconstructed in the correct geodetic coordinate system. In aerial photogrammetry, even if the drone itself is not RTK-enabled, accuracy improvements close to RTK-equipped systems can be achieved by measuring reference points beforehand with LRTK or by post-processing to attach precise coordinates to a subset of in-flight photos. Having high-precision geotags also streamlines SfM analysis itself: with more accurate initial values, the solver converges faster and more stably as a beneficial side effect.

An advantage of operating LRTK receivers in Japan is their compatibility with the Quasi-Zenith Satellite System “Michibiki” centimeter-class augmentation service (CLAS). This allows real-time reception of correction signals without setting up a dedicated base station on site, provided there is network coverage. By removing the need for expensive dedicated equipment, centimeter-level positioning has become far more accessible. With just a palm-sized LRTK device and a smartphone, high-precision positioning data can be collected anytime, anywhere—this convenience is revolutionizing photogrammetry workflows on site.

Practical workflow for aerial photography + LRTK × SfM: preparation → capture → processing → deliverables

Let’s review the steps for creating an SfM point cloud by combining drone aerial photogrammetry with LRTK. The process can be broadly divided into four steps: “preparation,” “capture,” “processing,” and “deliverable generation.”

• Preparation: Plan the flight area and prepare equipment. First determine the area, resolution, and altitude for drone capture and set up an automated flight plan. At the same time, attach the LRTK receiver to the smartphone, launch the app, and check connection status to GNSS base stations or network RTK services (or CLAS signals). If the drone is not RTK-equipped, it’s advisable to measure a few reference points on site to be used later for model alignment (LRTK can instantly provide known-point coordinates with centimeter accuracy).

• Capture: Execute the drone flight for aerial imaging. Following the automated flight plan and ensuring sufficient overlap, capture photos covering the entire site. If the drone supports RTK, high-precision coordinates will be recorded for each photo. Additionally, on the ground use LRTK Phone to scan and photograph areas that are difficult to capture by drone (such as spots in structural shadows or hidden slopes) while walking. Because blind spots that are hard to obtain from a drone can be supplemented by smartphone measurements, the resulting dataset becomes more complete. If high-precision geotags are attached to all aerial and ground photos acquired in this way, the subsequent processing becomes significantly easier.

• Processing: Import the captured photos into SfM analysis software (or a cloud-based point cloud service) to generate a 3D point cloud. The software computes camera positions and the point cloud through feature matching between images, but by using high-precision geotag information from LRTK, the model can be output aligned to real-world coordinates from the start. Use the reference points (GCPs) measured during preparation for accuracy verification or fine-tuning as needed. Nowadays, services exist that automatically generate point clouds by simply uploading photos to the cloud, so a high-spec PC is not always required. For example, LRTK Cloud can rapidly process large volumes of aerial images on the server side and generate high-precision point clouds and orthophotos in roughly an hour after upload, enabling same-day verification of processing results at the capture site.

• Deliverable generation: From the processing results you can obtain high-density 3D point clouds, wide-area orthomosaic images (aerial orthophotos), and various survey outputs such as digital elevation models (DEMs) and contour maps. You can also quickly create cross-sectional and longitudinal profiles from the point cloud or produce as-built drawings by comparing the data with design plans. With the LRTK × SfM workflow, the captured photos and point cloud data already contain accurate coordinate information, so traditional tasks like coordinate transformation or manual alignment become largely unnecessary. This enables rapid creation of up-to-date 3D models and various drawings, facilitating smooth integration into construction management tasks described next.

Heatmap differencing, quantity calculations, and construction quality control using point clouds

The high-precision 3D point clouds and orthophotos produced in this way can be used in many ways for construction as-built and quality control. Key use cases include the following features:

• As-built heatmap display: Overlay design data (design surfaces or planned 3D models) with the current SfM point cloud and visualize height differences as a color distribution (heatmap). By color-coding how far each point in the point cloud deviates from the design surface, you can instantly grasp areas of overfill, underfill, or finishing issues. For example, LRTK Cloud provides a function that overlays the design and point cloud and automatically colors design-compliant areas green, overfilled high areas red, and over-excavated low areas blue. This helps site supervisors intuitively understand as-built conditions across large areas and aids early detection of construction errors and reduction of rework.

• Quantity calculation (volume estimation): Volumes and quantities of earthworks can be quickly calculated from the acquired point cloud. Traditionally, calculating embankment or excavation volumes required generating cross sections from survey data and performing manual volume computations, but with a high-density point cloud you can automatically compute volumes for any selected area in the software. By calculating fill or excavation volumes from differences between point clouds, you can accurately quantify daily progress and quantities. LRTK Cloud can instantly measure earthwork volumes on the order of thousands of cubic meters, allowing field personnel to check on a tablet how many cubic meters remain to reach design elevations. This dramatically improves quantity management efficiency.

• Digital construction quality management using point clouds: Because SfM point clouds record the as-built condition in detail, they can be used to digitally verify construction quality against design values. For road works, for example, you can measure pavement gradients, widths, and heights on point cloud data to confirm compliance with design; for concrete structures, you can comprehensively check as-built dimensions. Since point clouds contain surface-wide information, areas that would previously have been assessed by interpolating between survey points can now be evaluated accurately. Keeping high-precision point clouds as inspection records also supports quality certification and future traceability. By leveraging point cloud data in this way, construction management PDCA cycles can be driven by data, improving the accuracy and reliability of quality control.

Operational efficiency and safety improvements through cloud sharing and real-time verification

Survey data generated by LRTK and SfM contributes to both operational efficiency and safety improvements through cloud sharing and real-time verification.

First, cloud-based data sharing speeds up information dissemination. Once point clouds and orthophotos captured on site are uploaded to the cloud, office staff and clients can instantly view the latest conditions. Since high-resolution 3D data are hosted server-side, recipients can inspect and measure them easily from a web browser without specialized software. This enables remote meetings and reporting that formerly required gathering on site, and reduces the need for high-cost workstations or point cloud viewers at each office, contributing to cost reductions.

Second, the ability to confirm and use data in real time has major safety management benefits. Drone aerial imaging allows safe observation of high, steep, or heavy-equipment areas that are difficult for workers to access. Remote sensing can replace previously hazardous surveying tasks, thereby reducing risks to personnel. Moreover, because LRTK enables immediate positioning and as-built checks on site, errors can be detected and corrected the same day. For example, if buried utilities are recorded in point clouds in advance and projected on site via AR, the risk of accidentally damaging underground pipes or cables during excavation can be avoided. When accurate information is shared in real time and everyone can intuitively grasp site conditions, both safety management and work efficiency improve dramatically.

Conclusion: The SfM surveying revolution powered by LRTK and the benefits of adoption

Combining RTK-GNSS technology with LRTK to improve and streamline photogrammetry (SfM processing) represents a true revolution in surveying workflows for field operations. It dramatically reduces the cumbersome burden of GCP installation while enabling rapid acquisition of centimeter-level 3D models comparable to laser scanners. As a result, surveying and as-built management tasks that once required many specialists and much manpower can now be performed routinely by fewer people in less time.

A primary benefit of adopting LRTK is the promotion of on-site DX (digital transformation). Using digital data consistently from surveying through construction management enhances visualization and automation of entire processes, reduces reliance on individual expertise, and standardizes operations. Another major advantage is labor reduction, boosting productivity and lowering costs. If one-person-per-device smartphone surveying becomes standard, even limited staff can cover expansive sites—offering a solution to growing labor shortages. Real-time sharing of accurate as-built data also helps prevent rework and promotes quality uniformity. In short, integrating LRTK raises site operations to the next level by balancing safety and efficiency.

Fortunately, these advanced technologies can be deployed on site without special large-scale equipment. For example, a smartphone and a drone are all you need to start simple photogrammetry with LRTK. While RTK-equipped drones and laser scanners used to be required, today anyone can achieve centimeter-level 3D surveying by attaching a palm-sized RTK receiver to a commercially available small drone and smartphone. Products such as LRTK Phone, which turns an iPhone into a versatile surveying device, have already appeared, making it increasingly possible for individuals to take measurements on site whenever they like. With lower initial investment and reduced requirements for specialized skills, now is an ideal time to consider starting high-precision photogrammetry with a smartphone + drone as the first step in on-site DX. You will likely be impressed by both the convenience and the quality of the deliverables.