Process Overview

Last updated on 2024-09-03 | Edit this page

To generate a photogrammetric model, we might choose to overlook the concepts and formulas utilised within the software to produce such a model.

Yet, understanding how the software works can be highly beneficial. Such understanding enables us to evaluate the feasibility of successfully acquiring objects and identify strategies to enhance our chances of success.

Basic steps

The basic steps of photogrammetric processing are:

  1. Feature detection
  2. Feature matching
  3. Structure reconstruction

Feature detection

Features are “interest points” or “key points” in an image.

The goal of this step is to find points which are repeatable and distinctive.

Corners and other distinctive patterns in the image are obvious features to consider.

sift algorithm
Feature points in photograph of building © Lukas Mach at English Wikipedia

Challenge: Try it yourself

Which points would you choose in this image of a building?

building
Édifice Shaughnessy, 401-407, rue McGill, Montréal © Jeangagnon at Wikimedia

Why?

Challenge: Try it yourself

Which points would you choose in this image of a vase?

vase
Vase by Jersey Glass Company of George Dummer (MET, 20.48.1) © Metropolitan Museum of Art at Wikimedia

What makes this image different from the one before?

Challenge: Try it yourself

Which points would you choose in this image of a silver mug?

mugs
Silver mugs created by Adrian Bancker (1703-1772) between 1731-1750. Currently held by the Museum of the City of New York. © Adrian Bancker at Wikimedia

Does this share the same challenges with any of the images above?

Feature matching

The goal of this step is to find correspondences of features across different views.

The software will attempt to match features in two or more images, ideally seeking a reliable outcome. Here we want a reliable result.

Features matched
View of features matched by Metashape photogrammetry software on a dataset of a coloured carboard box. Blue lines display valid matches, and red lines display invalid matches.

Challenge: Try it yourself

Do the features below correspond with each other?

hosue match

Structure reconstruction

Taking into account all identified features in a pair of images, the software builds a projection of the points in 3D space by using the camera position.

projection
Software computes 3D points describing the scene geometry © ZooFari in Wikimedia

The scene is incrementally extended by adding new images and triangulating new points.

point cloud
Point cloud projected by images

A much denser set of features is produced. The output of this process is a point cloud or a collection of points.

dense point cloud
Point cloud projected by images

The 3D model is created using what is know as triangulation. This process creates a 3D model with thousands or millions of triangles.

triangulated model
Triangulated point cloud projected by images
triangulated model
Zooming into the triangles of the 3D model

The texture is then mapped to this surface.

3D model with texture

Challenge: What does this all mean for me?

Based on what you have seen, reflect on what are the key things to consider when selecting objects for photogrammetry.

Which of the objects you work with would be the best candidates for photogrammetry and which would not be ideal?

Acquisition tips

  • Capture images with good texture.
  • Avoid completely texture-less, transparent and reflective images. The computer will have difficulty finding and matching features.
  • If the scene does not contain enough texture itself, you could place additional background objects, such as posters, newspapers, small objects etc.