SfM processing revolutionizing construction surveying: Generating precise 3D point clouds from photographs

In recent years, the surveying sites of construction and civil engineering have been paying great attention to "SfM processing," which generates 3D point clouds from photographs. Three-dimensional measurement of the current situation, which formerly required expensive surveying equipment and significant effort, can now be easily performed with digital cameras or drones that anyone can own, dramatically improving surveying efficiency and accuracy.

This article explains the mechanism and methods of SfM processing in an easy-to-understand manner, explores the differences and advantages compared to conventional surveying methods, and shows concrete applications to construction surveying such as as-built management, earthwork quantity calculations, and pre-construction topographic surveys. It also introduces the latest case studies combining drones, smartphones, and cloud services. In addition, the limits of SfM and comparisons with complementary technologies such as GNSS and LiDAR, as well as challenges and points to note when introducing SfM, are discussed in detail. Finally, we summarize an effective method to correct the accuracy of point cloud data obtained by SfM and unify coordinate references using simple surveying LRTK with a smartphone plus a high-precision GNSS terminal.

What is SfM processing (mechanism and methods)

SfM (Structure from Motion) is a method to reconstruct the three-dimensional structure (3D model) of an object or site from multiple photographic images. Originally a technique in the field of computer vision, it has recently been widely applied to photogrammetry in civil engineering and construction.

In SfM processing, distinctive feature points (high-contrast patterns, corners, etc.) are automatically detected across multiple overlapping photos, and matching common points allows simultaneous estimation of the camera positions and orientations at capture time and the 3D coordinates of each feature point. Based on these estimates, image information from multiple viewpoints is integrated to generate a high-density point cloud (detail reconstruction using the Multi-View Stereo method). Through this process, site topography and structures can be reproduced as precise 3D point clouds or polygon mesh models.

The basic workflow of SfM processing is as follows.

• Photograph the site from many different angles with numerous photos (ensure sufficient overlap between adjacent photos)

• Import the photographed image data into dedicated software and automatically detect feature points in each image

• Match corresponding feature points between photos and compute the camera positions and orientations (image alignment)

• Analyze parallax from the estimated camera positions and reconstruct a high-density 3D point cloud (point cloud generation)

• If necessary, create a 3D mesh model or orthophoto (top-down photographic map) from the point cloud

As described above, a major feature of SfM processing is that detailed three-dimensional data can be obtained by software processing without special equipment, as long as photos taken with a regular digital camera or drone are available.

Differences and advantages compared to conventional methods (accuracy, efficiency, cost)

With the advent of SfM processing, the following major differences and benefits have emerged compared to conventional surveying methods.

• Accuracy: With proper photography and processing, photogrammetry via SfM can produce 3D models with an accuracy on the order of several centimeters (centimeter-level accuracy, half-inch accuracy). This is sufficient for typical civil engineering survey tasks such as as-built management and earthwork quantity calculations. In terms of absolute accuracy at a single point, conventional total station surveys or high-precision GNSS positioning (RTK) can outperform SfM with millimeter-level accuracy in some cases. However, because SfM can measure wide areas in a surface manner, the consistency and overall accuracy of the data are very high, and by correcting with known points (ground control points) as needed, SfM can achieve accuracy comparable to conventional methods.

• Work efficiency and safety: Photogrammetry using SfM brings overwhelming efficiency improvements. For example, by flying a drone, aerial photography of a large development site can be completed in tens of minutes, and a single flight can acquire millions of survey points. Compared to traditional manual point-by-point surveying, on-site work time can be greatly reduced. Also, because surveyors do not need to enter hazardous slopes or roadways, safety is dramatically improved. Since any dimension can be measured later at a desk from the captured photos, the need to “return to the site for additional measurements” is reduced, improving overall operational efficiency.

• Cost: Required equipment is roughly just a camera or drone and a general-purpose PC, so initial investment can be kept low compared to purchasing and maintaining expensive equipment such as 3D laser scanners (LiDAR). Reducing outsourcing of survey work and completing data creation in-house can lead to long-term cost savings. Furthermore, digitally storing the current situation in detail once allows immediate reuse for design changes or additional construction, minimizing rework and additional surveying costs. SfM, which can provide high-density point clouds at low cost, is a powerful solution for sites with limited budgets or personnel.

Applications to construction surveying (as-built management, earthwork quantity calculation, pre-construction surveys, etc.)

High-precision 3D point cloud data created by SfM can be used in various surveying and measurement tasks in the construction field. Major application examples include the following.

• As-built management: The finished shape (as-built) of roads and development sites can be recorded and verified in 3D. By overlaying point cloud data with the design model, it is immediately apparent whether the slopes and heights of fills and cuts conform to the design, helping to detect nonconforming areas early. Whereas verification used to be done by thinning measurement points and checking cross-sections, point clouds enable surface-based and intuitive inspection.

• Earthwork quantity calculation: The volumes of material excavated or filled can be accurately calculated. For example, by creating point clouds of the ground surface before and after excavation using SfM and comparing them, cut and fill volumes can be determined with high accuracy. Undulations that are easily overlooked in conventional longitudinal and cross-sectional surveys are fully captured by point clouds, improving the accuracy of progress quantity calculations. Drone photogrammetry is also used to measure the volume of on-site spoil or material fill mounds.

• Pre-construction surveys: This use records current topography and surrounding environment in detail before construction begins. By aerially photographing a wide area with a drone, land boundaries, positions of existing structures, and tree proliferation can be preserved as a three-dimensional topographic map. This allows detailed understanding of the site during planning, aiding design changes and risk assessments. Also, saving a pre-construction terrain model enables post-construction comparisons for environmental impact assessment.

• Progress management: Regularly conducting photogrammetry on the site according to construction progress to visualize progress in 3D. For example, by photographing with a drone weekly and updating point cloud data, the progress of earthworks and the construction status of structures can be tracked over time. Because the entire site can be viewed from above, delays or mistakes in work can be discovered and corrected early, and the data can be used as materials for safety management.

These applications align well with the Ministry of Land, Infrastructure, Transport and Tourism’s promotion of i-Construction and CIM (Construction Information Modeling), and 3D point cloud-based site measurement is becoming standardized in many public works projects.

Integration with drones, smartphones, and cloud services

SfM processing can be used more efficiently and conveniently by combining it with various devices and services. Here are examples of integration with drones, smartphones, and cloud services.

• Point cloud generation by drone aerial photography: Drones are optimal for acquiring aerial photographs. Taking images from directly overhead enables the rapid creation of wide-area terrain models and orthophotos. In particular, using RTK-equipped drones records position coordinates with high accuracy, allowing output point clouds to be aligned with the survey coordinate system from the start. This reduces the effort of placing many ground control targets, and only a few points for accuracy verification are sufficient. Drone photogrammetry is spreading across a wide range of uses, such as understanding the topography of development sites and infrastructure inspection, and is becoming a standard method on i-Construction sites.

• 3D measurement using smartphones: With improvements in current smartphone camera performance, everyday smartphones can easily capture photos for SfM. For small structures or indoor spaces, photographing from all sides with a smartphone alone can produce a 3D model. In addition, the latest iPhones and iPads have built-in LiDAR sensors that, with dedicated apps, can capture simple point clouds in real time. Furthermore, by combining a smartphone with a high-precision GNSS receiver (as in the LRTK described later), accurate position information can be attached to the photos for use in SfM processing. On-site, smartphone-based ground photography can supplement drone photos in blind spots that are difficult for drones to capture (such as under bridges or beneath tree canopies), and the ground-based photos can be merged with drone-generated point clouds. Smartphone + SfM is a cost-effective and user-friendly method suitable for small-to-medium sites or urgent measurements.

• Combining with cloud services: SfM processing involves heavy computation handling many image files, but cloud-based point cloud generation services have been increasing. By simply uploading photo data via the web, 3D point clouds and orthophotos are automatically generated, so a high-spec PC is not required. Point clouds generated in the cloud can be shared with stakeholders over the internet or displayed and measured in browser-based 3D viewers. If photos are taken on site and immediately processed in the cloud, completed point cloud models can be reviewed by the time you return to the office, enabling speedy use. Even without in-house specialists, cloud services allow advanced SfM analysis, lowering the barrier to technology adoption.

Limits of SfM and complementary technologies (comparison with GNSS, LiDAR)

Although SfM processing is convenient, there are cases where its application is difficult and technical limitations exist. Therefore, GNSS positioning technology and LiDAR (laser scanners) are often combined to complement the weaknesses of SfM.

• Limits of photogrammetry: Because SfM depends on optical photography, surfaces without features or made of highly reflective materials do not yield sufficient point clouds (examples: plain white walls, water surfaces, mirror surfaces, nighttime photography). In areas with dense trees, aerial photos may not capture the ground surface well, resulting in gaps in the point cloud. Parts not visible in images (the back sides of objects or shaded areas) cannot be obtained, and even with shooting from multiple directions, blind spots may remain, leaving holes in the model. Furthermore, 3D models obtained by SfM are basically in a relative coordinate system, so some reference information is needed to provide scale (dimensions) and absolute position. Aerial photography with unmanned aircraft is also affected by weather, and drone flights in strong winds or shooting in rain are difficult. Thus, SfM alone cannot handle all situations.

• Correcting coordinates and scale with GNSS: Combining GNSS (Global Navigation Satellite System) positioning with SfM can solve the aforementioned coordinate reference issue. If you obtain the shooting positions of photos or coordinates of targets using high-precision GNSS (RTK or PPK), you can scale and align the SfM-generated point cloud model to those known coordinates. This allows the 3D point cloud to have absolute position information in national geodetic or site coordinate systems. By verifying errors with control points, the reliability of SfM models can also be ensured. For example, installing a few RTK-GNSS observation points on site enables integration of SfM results into existing survey coordinates with an accuracy of several centimeters (cm level accuracy, half-inch accuracy), realizing precision control that would be difficult with SfM alone.

• Comparison and division of use with LiDAR: LiDAR (Light Detection and Ranging) measures distances directly to targets with laser light to create point clouds. LiDAR surveying can measure at night or in dark environments, capture shapes of textureless objects, and handle multiple reflections that can detect ground through foliage. Accuracy is also high, and high-performance equipment can achieve millimeter-level precision. On the other hand, LiDAR equipment is expensive and requires specialized handling, so it is less advantageous in cost and convenience compared to widely available camera-based SfM. Also, LiDAR point clouds are essentially monochrome point collections, and color information or the clear orthophotos offered by photographs are not obtained. Therefore, in practice, the two are often used together: typically, the majority of the site is measured with the easy SfM approach, and only areas that SfM cannot capture are supplemented with LiDAR. Such hybrid measurement enables construction of digital models that balance cost, accuracy, and completeness.

Challenges and points to note when introducing SfM (accuracy verification, shooting conditions, point cloud processing)

When newly introducing SfM photogrammetry to the field, there are several issues to keep in mind. Below are summarized methods for accuracy verification, shooting considerations, and points for point cloud data processing.

• Accuracy verification: When using SfM for the first time or for important measurements, verifying accuracy by combining with conventional methods is essential. For example, survey several locations on site with a total station or RTK-GNSS and compare the coordinates of the same points on the SfM-derived point cloud to quantitatively evaluate errors. Also, compare software-reconstructed reference scales (for example, a known-length ruler or grid) with actual measurements to check for scale errors in the model. Through these verification tasks, confirm that SfM results meet required accuracy, and if insufficient, perform additional photography or introduce more control points for correction.

• Shooting condition points: High-quality photos are key to SfM accuracy. Ensure sufficient overlap (over 70% between adjacent photos) and photograph subjects from various angles. To obtain sharp images, avoid camera shake and subject blur by using tripods or drone auto-flight functions as needed. Shoot during bright daytime hours and avoid extreme backlighting or strong shadows to improve feature point extraction accuracy. For large sites, divide the area and photograph sequentially to avoid omissions, and keep camera focal length and resolution consistent to stabilize later analysis. It is safe to take several test shots and perform a quality check with the software beforehand to confirm there are no reflections or omissions before the main shooting.

• Point cloud processing and data use: Point clouds generated from photos can be extremely large (tens of millions of points or more), so appropriate processing and management are necessary. Remove isolated noise-like points or obviously misplaced points using filtering functions. If the survey target is ground surface measurement, remove unnecessary objects such as trees or vehicles or perform automatic extraction (classification) to isolate only the ground. When point clouds are difficult to handle as-is, create mesh models, generate contour lines, or extract required cross-sections to produce deliverables. Also consider cloud storage or dedicated viewers for sharing because point cloud file sizes can be large. Training in point cloud processing skills within the company and trials to master operations are important at the early stage of introduction. With these precautions, SfM photogrammetry can be smoothly applied to field operations.

Summary: Correcting SfM accuracy and unifying coordinate references with LRTK (smartphone + high-precision GNSS)

SfM processing, which can generate precise 3D point clouds from photographs, has brought great innovation to construction surveying. As described in this article, SfM alone is extremely useful for improving field measurement efficiency, but its value is further enhanced by combining it with other technologies.

Of particular interest is the use of LRTK, which integrates a smartphone and a high-precision GNSS. LRTK is a small RTK-GNSS receiver that can be attached to a smartphone, enabling centimeter-level positioning on site with ease. By using LRTK during photography, the position information recorded for each photo can be highly accurate, providing the SfM-generated point cloud with a definite scale and coordinate reference. Tasks that previously relied on placing targets or surveying known points for SfM alignment can be greatly simplified by using LRTK.

Moreover, LRTK allows supplemental surveying of detailed points not included in the SfM point cloud and on-the-spot measurement of verification points for as-built drawings, enabling immediate on-site feedback. Because the smartphone + GNSS combination is easy to use, a single person can perform shooting and positioning concurrently while moving around the site, allowing efficient surveying with limited personnel.

Combining SfM photogrammetry with LRTK positioning realizes an ideal workflow of “quickly 3D-mapping wide areas while placing the whole dataset on accurate coordinates.” In future construction sites, this kind of measurement DX centered on smartphones will further advance. Combine innovative SfM techniques with practical tools like LRTK to help improve the productivity of your daily surveying operations.