Introduction to SfM Processing Software

What Is Structure from Motion?

Structure from Motion (SfM) is a photogrammetric technique that reconstructs three-dimensional geometry from a series of overlapping two-dimensional photographs. The software analyzes common features visible in multiple images, calculates where each camera was positioned and aimed when the photo was taken, and then triangulates the position of thousands of matching points in three-dimensional space. The result is a sparse point cloud: an initial set of 3-D coordinates that represents the geometry of the scene and confirms that the photo alignment worked correctly.

Agisoft Metashape is one of the most widely used SfM applications in geospatial and surveying workflows. The steps covered in this lesson—adding photos, aligning them, and inspecting the sparse cloud—form the foundation of every photogrammetric project, whether you are mapping a construction site from a drone or documenting a historic facade from the ground.

The SfM Workflow at a Glance

A typical Metashape project follows this sequence:

1. Add Photos – import the image files into a chunk (Metashape's term for a processing group).

2. Align Photos – detect keypoints, match them across images, and solve for camera positions and orientations.

3. Build Dense Cloud – generate a much denser point cloud from the aligned cameras (a later lesson).

4. Build Mesh or DEM – create surface models (a later lesson).

5. Export products – deliver orthomosaics, DEMs, or point clouds to clients.

This lesson focuses on steps 1 and 2 only.

Adding Photos and Organizing a Chunk

When you open Metashape and select Workflow → Add Photos, you are telling the software which image files belong to this project. Keep all images for one flight or one capture session in a single chunk unless you have a specific reason to separate them (for example, combining nadir and oblique captures). Metashape reads EXIF metadata embedded in JPEG or TIFF files; GPS-tagged images will appear as approximate camera positions on the reference pane even before alignment.

Before running any processing, verify that your images load correctly: check that thumbnail previews appear, that EXIF coordinates populate in the Reference pane, and that there are no corrupted files (shown with a warning icon).

Aligning Photos: Key Parameters

The Align Photos dialog contains several settings that directly affect output quality and processing time.

Accuracy

The Accuracy setting controls the resolution at which the software searches for matching keypoints:

• Highest / High – uses full-resolution or near-full-resolution images. Finds the most tie points and produces the most accurate camera positions. Processing time is significantly longer.
• Medium – downscales images before matching. A useful compromise when computing time is limited.
• Low / Lowest – fastest, but tie-point count drops sharply and alignment failures become more likely.

For professional survey deliverables, High accuracy is the standard starting point.

Key Point Limit and Tie Point Limit

• Key Point Limit caps how many candidate feature points the detector extracts from each image. A higher limit gives the matcher more opportunities to find correspondences but increases memory use.
• Tie Point Limit caps how many matched points per image are kept after matching. Setting it to zero (0) in Metashape means no limit, retaining all valid matches.

Generic and Reference Preselection

• Generic Preselection compares lower-resolution versions of images first to identify which pairs are likely to overlap, speeding up matching.
• Reference Preselection uses GPS/EXIF coordinates to select candidate image pairs. It is faster than Generic but requires reliable geotag data.

Evaluating the Sparse Cloud

After alignment, examine three indicators:

1. Tie-point count – visible in the model view tooltip or the report. More tie points generally mean more robust geometry, though redundant or low-quality points can add noise.

2. Reprojection error (RMS) – measures how far each tie point's projected position deviates from its observed position across images, expressed in pixels. Values below 1 pixel are typically acceptable; values above 2–3 pixels suggest poor image quality, incorrect camera calibration, or mismatched features.

3. Camera alignment coverage – aligned cameras appear as blue frustum icons; unaligned cameras appear gray or are flagged with a warning. A cluster of unaligned cameras usually signals a low-overlap area or blurry imagery.

Diagnosing Low-Overlap Problems

Insufficient image overlap is one of the most common field mistakes. When a portion of the project area was captured with fewer than the recommended number of overlapping images (typically 75 % forward overlap and 60 % side overlap for mapping missions), the software cannot find enough common features, and cameras in that area fail to align. In the camera position view, you will see a gap—a region with no blue frustum icons—corresponding to that area on the ground.

Solutions include:

• Re-flying the gap area with corrected overlap settings.
• Adding oblique imagery to introduce new viewing angles.
• Using ground control points (GCPs) near the gap to constrain geometry.

For future missions, plan flight lines carefully using your ground sampling distance (GSD) requirements and overlap settings in your flight planning app before takeoff.

Safety and Professional Conduct

During computer lab sessions, standard workstation safety applies: keep liquids away from equipment, save work frequently, and follow instructor procedures for logging off shared machines. When discussing datasets collected in the field, remember that drone imagery may include sensitive infrastructure; handle data according to your organization's data security policy. Never share client or training datasets outside the class without authorization.