Basic Terms Beginners Should Know About SfM Processing: Understand the Basics of Photogrammetry

Introduction

Recently, in construction, civil engineering, and surveying fields, you may increasingly hear terms like SfM processing and photogrammetry. This technique, which creates 3D models on a computer from site photos taken by drones or digital cameras, is attracting attention because it can improve work efficiency and safety. However, for those encountering SfM processing for the first time, it can be confusing: “What does it mean to create a 3D model from photos?” or “There are so many technical terms—it looks difficult…”

This article gently explains the basic technical terms that beginners in SfM processing should know. Organizing the basic terms of photogrammetry will make it easier to understand how the technology works and will help with communication and learning in your work. Technical terms are explained as simply as possible, so if you are about to start drone-based photogrammetry, are working to apply ICT-based new surveying methods in the construction industry, or are a local government employee with no prior surveying experience, please read on with confidence.

Basic Technical Term Explanations

First, let’s check the basic technical terms frequently used in the fields of SfM processing and photogrammetry.

• SfM (Structure from Motion): A computer vision technique that reconstructs the three-dimensional structure (shape) of an object from multiple photographic images. Software analyzes a set of photos taken from different positions and angles and reconstructs the shapes of the objects appearing in them as 3D models. Because it can automatically create 3D models from photos taken by drones or cameras, it is a central method in photogrammetry.

• Photogrammetry: A general term for methods that use photographic images to measure the dimensions and shapes of terrain and structures. Historically developed as a technique to create maps from aerial photographs, advances in digital imaging and SfM processing now allow anyone to generate detailed 3D data from photos. Numerous photos from drone aerial surveys or ground photography are overlapped and analyzed to produce point cloud data and 3D models of targets.

• Point Cloud Data (Point Cloud): 3D data that represents an object’s shape using many “points” in space. Each point has coordinates (X, Y, Z) and indicates a position on the surface of the object. In photogrammetry, software generates many points by matching features across photos, reproducing shapes like buildings and terrain as collections of points. The higher the point density, the more detailed the shape capture; if the point cloud is dense enough, it can represent the subject with an appearance close to the photos.

• Mesh: A 3D model composed of polygons generated from point clouds. While point clouds treat countless points directly, mesh models create continuous surfaces by connecting points to form faces (polygons). Usually constructed as collections of triangular polygons and automatically generated from point cloud data, meshing makes the 3D model easier to handle and allows textures to be applied to the surface for a more realistic appearance.

Key Terms Used in the Processing Flow

In the processing flow where photogrammetry software generates 3D models from photos, specialized terms appear at each step. Here are the main terms you should know in the SfM processing workflow.

• Feature Points: Distinctive points in an image (areas with strong contrast, corners, patterns, etc.). SfM automatically detects many feature points from each photo. These serve as markers to find common areas across photos and form the basis for later 3D reconstruction.

• Matching (Feature Point Correspondence): The process of associating the same feature points appearing in multiple photos. Software compares feature point patterns between photos to find corresponding points on the same object. With a sufficient number of matches, it becomes possible to estimate the relative positions from which each photo was taken.

• Bundle Adjustment: A computational process that simultaneously optimizes camera (photo) parameters and point cloud positions based on the feature point matching results. For all photos, it adjusts the camera positions and orientations at the time of capture (intrinsic and extrinsic parameters) and the coordinates of the reconstructed 3D points to minimize errors. Simply put, it is a fine-tuning step that slightly moves the positions of the photos and the 3D points so that “the positions of the observed feature points align consistently.” This process determines the accurate position and orientation of each photo and produces an initial 3D point cloud.

• Sparse Point Cloud: A sparse point cloud obtained as a result of bundle adjustment. The positions of the feature points used for matching are calculated by triangulation, generating a relatively sparse set of points. Because it is a sparse point cloud, it only indicates the rough shape of the object, but it represents a stage where camera arrangement and scale consistency have been established.

• Dense Point Cloud: A much higher-density point cloud than the sparse point cloud. After camera positions are fixed, the software estimates depth in detail for each image and generates point clouds that cover nearly the entire surface of the object. The resulting dense point cloud is a detailed 3D dataset with points representing fine details of buildings and terrain. Generating a dense point cloud prepares the foundation for subsequent mesh creation and texture mapping for precise 3D modeling.

Terms Related to Accuracy and Coordinates

To use 3D data produced by photogrammetry for surveying results or design, it is important to improve the model’s accuracy and align it with real-world coordinate systems. Here are the main terms related to accuracy management and coordinate settings.

• GCP (Ground Control Point): Reference points placed on the ground whose accurate coordinates have been measured in advance. In photogrammetry, multiple GCPs are placed within the target area and their positions are associated with the photos in the software to give the generated point cloud and model the correct real-world coordinates and scale. Using GCPs can improve the overall positional and elevation accuracy of the 3D model.

• RTK (Real-Time Kinematic): A technology that uses GNSS (satellite positioning) to perform centimeter-level (half-inch-level) high-precision positioning in real time. By sharing satellite data between a base station (fixed receiver) and a rover (measurement receiver) and correcting errors, it provides much more precise positional information than ordinary GPS. In photogrammetry, RTK-equipped drones can tag each photo with high-precision coordinates (geotags), or GCP coordinates on site can be obtained by RTK surveying, ensuring high spatial accuracy of models.

• Georeferencing: The process of tying created point cloud data or models to a real-world geodetic coordinate system (position alignment). In simple terms, it is the work of aligning the completed 3D model to the correct position and scale on a map. In photogrammetry, point clouds are aligned to real-world coordinate systems using known GCP coordinates or photo geotag information. Georeferenced data can be overlaid with other map data or CAD drawings.

• Scale Bar: A known-length scale or marker placed in the photos. For example, by placing a ruler or a rod-shaped marker with a precisely known length near the subject and entering the actual distance between two points into the software, the model’s scale can be set correctly. If GCPs cannot be installed, using a scale bar allows the model generated by photogrammetry to be matched to real-world dimensions.

• Geotag: Positional information added to a photo (latitude, longitude, altitude of the shooting location, etc.). When taken by a drone or smartphone, coordinates of the shooting location are automatically recorded in each image’s EXIF data. Photogrammetry software reads these geotags to set initial shooting positions for each photo. Using geotagged photos allows the reconstruction to start with the model’s position and orientation somewhat aligned to the real world, making later alignment tasks easier.

Output Model Related Terms

Also familiarize yourself with terms related to deliverables (output data) generated by photogrammetry software. Below are representative data types output as 3D models or images and what they contain.

• Orthoimage: A composite photograph (orthophoto) created from a top-down viewpoint. Multiple aerial photos are stitched together and their distortions corrected to create an integrated image from a nadir view like a map. Because every part is at the same scale, distances and areas can be accurately measured on an orthoimage. In drone photogrammetry, an ortho mosaic image of the entire site can be produced from the generated 3D model and used to grasp the current situation like a plan view.

• Texture Mapping: The process of applying photographic images to the surface of a 3D mesh model. Mesh shapes obtained by photogrammetry contain only geometry, but by overlaying the corresponding photos’ colors and patterns onto the surface, the model can be made to look very much like the real thing. Textured models have more visual information than point clouds or untextured meshes and are suitable for intuitively understanding the site.

• 3D Mesh (Polygon Model): A three-dimensional surface model generated from a point cloud. It represents the object surface by combining many small triangular polygons, producing a smoother and more continuous shape than point cloud data. 3D mesh models obtained from photogrammetry reproduce actual contours and shapes of buildings and terrain and can be used for design, simulation, VR visualization, and more.

• Polygon: Each face element (polygon) that makes up a 3D mesh. Models created by SfM typically use countless triangular polygons. The model’s level of detail and data size are determined by the number of polygons; models with more polygons represent finer details more smoothly but have larger file sizes and require heavier processing. Therefore, models are often simplified (reduced) to an appropriate polygon count depending on the use.

Other Related Terms

Finally, let’s briefly touch on other related terms surrounding SfM photogrammetry.

• Cloud Processing: Services or methods that perform photogrammetry data processing on servers in the cloud. Generating 3D models from numerous photos requires high-performance computers, but using cloud processing allows users to upload photo data over the internet and have the server automatically generate point clouds and models. Users only need to download the result data, making it easy to request processing from the field.

• Image Overlap (Overlap Ratio): The proportion of overlap between captured photos. Low overlap reduces the number of corresponding feature points between photos and hinders successful 3D reconstruction. Therefore, photogrammetry recommends ensuring sufficient overlap between adjacent photos (for example, about 80% in the forward direction and 60% or more in the lateral direction). When shooting a site with a drone, it is important to set flight routes and shooting intervals that account for required overlap.

• Flight Path: The route that the camera (drone) travels during shooting. Flight paths are planned in advance to comprehensively capture the subject from various angles. For example, when surveying a wide area, a parallel zigzag route may be flown, while for structures the drone may circle around to capture them from all sides. Obtaining sufficient photos along a stable flight path reduces the likelihood of holes or gaps in the 3D model after SfM processing.

Conclusion

Above, we introduced the main basic terms used in SfM processing. At first, the technical terms may seem numerous and difficult, but by understanding the meaning of each word, the flow and mechanism of photogrammetry become clearer. Grasping these basic terms will greatly help communication when handling 3D models on site and in learning related technologies.

In recent years, solutions have emerged that leverage these technical elements behind the scenes to enable anyone to perform three-dimensional surveying easily. For example, simple surveying using LRTK combines a smartphone with high-precision GNSS to capture a site and instantly obtain a high-precision 3D point cloud model with global coordinates. This leverages the technologies of RTK and SfM discussed in this article behind the scenes so that users can get results without being conscious of technical terms or complex procedures. By using such tools, even beginners can quickly and safely perform three-dimensional recording and surveying of sites. Deepen your understanding of the basic terms and take a step toward incorporating new technologies into your field work.