What Is SfM Processing? A Beginner’s Guide to the Technology That Generates 3D Models from Photos

The use of 3D models is spreading in the construction industry and municipal operations. One technology attracting attention for generating 3D models from photos is SfM (Structure from Motion). This article explains, in an easy-to-understand way for beginners, how SfM processing works, the basic steps, required equipment, differences from other surveying technologies, benefits of implementation, and points to watch. Finally, it also introduces how to utilize SfM in combination with simple smartphone surveying using LRTK.

1. What Is SfM Processing?

SfM is a technique that reconstructs three-dimensional models from multiple photographic images. Without using specialized surveying instruments, it can reconstruct the 3D shape of objects or terrain from multiple photos taken with ordinary cameras or drones, making it a form of photogrammetry that has attracted attention. Originally developed in the computer vision field, recent advances in software have expanded its use in construction and civil engineering.

Advantages of SfM include the ability to start at low cost because expensive laser scanners and other specialist equipment are not required; good compatibility with drones, enabling quick measurement over wide areas; and the ability to obtain detailed, colorized 3D models from high-resolution photos. The analysis can automatically generate outputs such as point clouds, 3D meshes, and orthophotos (overhead photographic maps). On the other hand, results are heavily influenced by the quality of the captured photos, and processing large numbers of photos can require long computation times. Still, because it balances ease of use with improving accuracy, SfM is expected to support on-site DX (digital transformation).

2. Process from Photos to a 3D Model

SfM processing involves several steps from photographing to obtaining a 3D model. Let’s look at the basic flow step by step.

• Photo capture: First, take photos of the object or area from various angles. A key point is to capture images so that adjacent photos overlap by at least 80% (forward/backward) and 70% (side-to-side). When using a drone, plan automatic flights to capture images systematically, and for three-dimensional objects such as buildings, also take oblique shots to capture side information. Take photos as brightly and sharply as possible, avoiding blur and uneven exposure.

• Feature point extraction: When you load the photos into dedicated software, the software first automatically detects landmark points called feature points in each photo. Many distinctive points are extracted—high-contrast patterns, edges, corners—that can be recognized despite viewpoint changes.

• Feature point matching: The extracted feature points are compared across photos to pair points that likely represent the same location. By finding sets of feature points that appear in multiple images, the software identifies which photos overlap. Since incorrect correspondences (mismatches) can occur, processing to remove outliers using algorithms like RANSAC is also performed.

• 3D reconstruction (structure estimation): Based on corresponding points across photos, the camera positions and orientations (external parameters) and the 3D coordinates of the feature points are calculated simultaneously. This is achieved through triangulation (measuring distances using angle differences from multiple viewpoints), where the spatial coordinates of a point are derived from the positional relationships of the photos in which the point appears. At this stage, you obtain the camera poses for each photo and a sparse point cloud representing the object. Then, using the estimated camera positions and the sparse point cloud, a method called Multi-View Stereo is applied to densify the point cloud. Depth is estimated for pixels in each image, gaps are filled, and dense point clouds or 3D mesh models are generated. Finally, integrated point cloud data and 3D models are output.

3. Required Equipment and Software

Here is an explanation of the equipment needed for photographing and the software used in SfM processing. Special surveying instruments are not necessary—common cameras and drones can be used.

• Camera: A high-resolution digital camera is preferable. DSLRs and mirrorless cameras are ideal, but compact digital cameras can also be used. Fixing focal length (zoom) and focus during shooting so the field of view does not change improves accuracy in post-processing.

• Drone: Aerial drones are ideal for large-area photogrammetry. By automatically acquiring numerous photos from the air, you can efficiently create 3D models of terrain and large structures. Recently, drones equipped with RTK-GNSS have become more common, allowing high-precision geotagging of captured images.

• Smartphone: Modern smartphones have strong camera capabilities and can be used to create 3D models of small objects. While a smartphone alone can capture photos for SfM, adding dedicated apps or accessories to embed positional information can allow use in surveying applications.

• RTK positioning equipment: To give an SfM-generated model absolute coordinates (real-world surveying coordinates), control points (GCPs) or high-precision GNSS positioning are required. RTK-capable GNSS receivers are available for this purpose. RTK (Real Time Kinematic) is a technique that enhances satellite positioning accuracy and can achieve centimeter-level precision. For example, smartphone-compatible RTK systems like LRTK can add high-precision position information to captured photos, making it easier to generate 3D models in an accurate coordinate system without installing many ground control points.

Software: Dedicated software is required for SfM processing. A range of tools is available from commercial to free. Open-source options like Meshroom and COLMAP are usable, but they can be difficult for beginners to configure, so starting with user-friendly commercial software or cloud services is often recommended. Recently, cloud services that automatically process images after uploading have appeared, enabling large-scale SfM analysis even without a high-performance PC.

4. Differences Between SfM and Other Technologies

SfM is a convenient technology, but it’s important to understand its characteristics compared with other surveying and 3D measurement methods. Below are differences when compared to laser scanning, BIM, and drone LiDAR.

• Difference from laser surveying: Laser measurement (3D scanning) is an active method that uses LiDAR and similar devices to emit laser pulses at objects and directly measure distance from their reflections. SfM, on the other hand, is a passive method that infers shape from images captured by a camera. Laser scanning can operate in dark environments and capture shapes of textureless objects (for example, a uniformly white wall), while SfM cannot generate point clouds if features are not visible in photos. Conversely, SfM has lower equipment costs and is easy to operate, and can measure wherever a camera can reach. Laser scanners are expensive and require expertise, but SfM can often substitute using a familiar camera.

• Difference from BIM: BIM (Building Information Modeling) refers to digital design documents and 3D models used in architecture and civil engineering. Unlike SfM, which is a capture-based measurement technology, BIM is a human-created design information model from the planning stage. BIM models include attributes and dimensions of components, whereas point clouds and meshes from SfM are shape-only data. However, SfM-generated point clouds can be combined with BIM as an as-built 3D model. For example, you can create a BIM model that reflects the existing condition based on SfM point clouds during renovations, or overlay point clouds acquired during construction onto design BIM to check as-built conformity.

• Difference from drone LiDAR: Drone LiDAR mounts a lightweight laser scanner on a drone to measure terrain and structures from the air. Although its purpose is similar to drone photogrammetry using SfM, the data acquisition methods differ. LiDAR directly measures distance with lasers and can capture ground points through gaps in tree canopies or acquire point clouds for low-feature terrain, which is its strength. Some systems can generate point clouds in real time during flight. However, LiDAR equipment is very expensive and complex to handle, and the point clouds obtained are monochrome intensity data. If you need high-resolution, color 3D models, SfM is more suitable. For large-area terrain surveys, SfM is cost-effective, while LiDAR is preferred when vegetation-covered ground or complex targets such as power lines are involved.

5. Benefits of Implementing SfM in Construction, Civil Engineering, and Municipal Work

What benefits can you expect by implementing SfM at construction sites, civil engineering works, or in municipal operations? Here are the main advantages.

• Low cost: You can significantly reduce initial investment costs by avoiding the purchase of dedicated surveying instruments or laser scanners. Existing digital cameras and drones can be used, and software options range from free to commercial. Tasks that once required costly equipment have become much more accessible with SfM.

• Ease of use: Even without specialized surveying knowledge, you can start photogrammetric 3D surveying simply by taking photos and processing them with dedicated software. Single-person operation is possible, reducing staffing needs. Many GUI-driven tools are intuitive, making them easy for junior engineers and beginners to adopt.

• Accuracy: While some may assume photogrammetry is low-accuracy, with the right approach you can achieve errors on the order of a few centimeters. By installing ground control points (GCPs) and aligning the model to known coordinates, you can attain accuracy comparable to traditional terrestrial surveys. Repeated measurements yield stable results, making SfM useful for as-built verification and displacement measurement.

• Time savings: Because large areas can be photographed quickly, substantial time savings are possible. For instance, tasks that previously required point-by-point measurement by personnel have been reported to be completed in about half a day with drone aerial photography. Processing after capture takes time, but on-site workdays can be greatly reduced. A single flight can capture millions of points, reducing the need for re-measurement and streamlining the overall workflow.

• Safety: Data can be collected remotely even in hazardous locations, improving worker safety. Slopes at risk of collapse, disaster sites, and aging infrastructure that are dangerous to approach can be surveyed by drone or from a safe distance. Reducing the need for scaffolding or climbing lowers accident risk while still allowing measurements.

• Digital utilization (DX promotion): Point clouds and 3D models obtained from SfM are digital assets that can be imported into CAD or GIS for analysis and sharing. Information that was hard to grasp in 2D drawings or photos becomes intuitive in a 3D model, improving on-site information sharing and decision-making and enabling advanced construction and maintenance management. SfM contributes to DX (digital transformation) in these operations.

6. Common Misconceptions and Points to Watch

When performing SfM processing, beginners often fall into certain misconceptions or overlook important considerations. Below are representative points explained.

• Importance of GCPs (control points): To align an SfM model to a real-world coordinate system, you need ground control points (GCPs). Photos alone can provide relative 3D shape, but without GCPs the model’s scale and position remain arbitrary. By measuring known-coordinate points onsite and specifying them during processing, you give the model accurate scale and absolute position. Without GCPs, absolute accuracy is not guaranteed, and errors can accumulate over large areas. For tasks requiring high precision, always place a sufficient number of GCPs or complement with RTK-capable drones.

• Coordinate system settings: To overlay SfM-derived point clouds or models with maps or design data, you must unify coordinate systems. Models from drone photos are by default in the camera coordinate system (an arbitrary axis) and will not match maps as-is. Use GCPs or RTK positioning to convert models to public coordinate systems (such as Japan’s plane rectangular coordinate system or global geodetic systems). Neglecting coordinate transformation can render 3D models unusable with other maps or drawings.

• Processing time and PC specifications: SfM processing involves heavy image analysis and requires considerable time and computing resources. Processing hundreds of photos can take several hours to half a day even on high-performance PCs. The load increases with image resolution and number of photos. It is difficult to get immediate results on site, so plan with sufficient time from surveying to deliverables. Consider using cloud processing services or running batch jobs overnight to make efficient use of time.

• Shooting conditions (weather and reflections): Pay attention to environmental conditions during photography. Slightly overcast weather, which may seem dim, is actually good because shadows are reduced and light is even (rain is not suitable). In contrast, clear midday sun can create strong direct light and large brightness differences between photos, negatively affecting matching. Avoid extreme shadows by choosing appropriate times and exposure settings. Highly reflective surfaces like water or glass can disrupt feature detection and matching due to reflections. For such areas, try changing shooting angles or using polarizing filters if possible to reduce reflections. Even then, reconstruction may be difficult, and some parts may end up missing from the point cloud.

7. Conclusion

SfM processing is a groundbreaking technology that makes it easy to generate 3D models from photos. Even beginners can obtain practically usable 3D data at job sites if they understand the key points and follow the correct procedures. Using this technology in construction sites, civil works, and municipal operations not only improves efficiency and safety but also greatly contributes to the digitalization (DX) of tasks.

In particular, the combination with simple smartphone surveying using LRTK is excellent. By combining high-precision RTK positioning with a smartphone for photogrammetry, anyone on site can easily acquire 3D models with absolute coordinates. Smartphones plus LRTK can complement drone limitations by capturing details and blind spots that are hard to photograph with drones. By adopting such new approaches, surveying tasks that used to require specialists can be performed by on-site personnel, significantly lowering the barriers to technology adoption. Take advantage of the combination of SfM and smartphone surveying to help drive DX at your sites.