Structure from Motion

Rendering of a SfM setup Here, we will talk about Structure from Motion, a technique that enables you to create 3D objects out of 2D photos. The information is divided into different chapters, as we sequentially progress within our course. The purpose of this resource is to prepare you for each course as part of your time for self-study (Selbststudium) as well as to operate as a general reference for the remainder of the term. It will not replace the content of the course, rather then be an additional resource. The content of this Wiki is not equal to the content of the course!

Everything here is written in English and as always work in progress. As this is no traditional Wiki, where you are able to participate, I still would enjoy your comments on this online resource. Typos, broken links, unclear passages… Let me know, if I can improve this page and how to do it, it will be much appreciated.

Taking the photos
Capturing Photos Chapter 1: Taking the photos
Here, you will learn how to take photos for SfM. It is a simple-enough process, but there are some points to consider. The equipment can be very basic (a simple camera or smartphone) to pretty elaborate (DSLR, Colour Chart, Coded Measures). However, the process is basically the same.
Estimated reading time: 5 minutes
Adobe Bridge Screenshot Chapter 2: Post-processing the photos in 🪟 Windows | 🐧 Linux
Taking the photos is just the first step. We also need to color-correct them and take care of possible lens distortions. We usually use Adobe Bridge and Camera RAW (Windows) for that, although these two software are not free. A free alternative under Linux is darktable.
Estimated reading time: 2 minutes
Creating the 3D object
metashapedownload.jpg Installing Metashape
We use Metashape for our purposes. If you work from home, you can test the software for 30 days. Here is how.
Estimated reading time: 1 minute
After adjusting and positioning, your object should be in the centre and... Chapter 3: Feature Detection
The first step towards a 3D object is to figure out where the features are and in what relation they are standing together.
Estimated reading time: 5 minutes
sfm_dense_clean.jpg Chapter 4: Bundle Adjustment
A low-density point cloud looks nice, but you what looks even better? A high-density point cloud.
Estimated reading time: 3 minutes
sfm_mesh_clean.jpg Chapter 5: Meshing the points
We still can see though our model, so we have to change that. Meshing connects points to polygons and polygons are solid.
Estimated reading time: 4.5 minutes
sfm_finished.jpg Chapter 6: Texturing the 3D model
Having the mesh ready, we still need some paint. With texturing, we will project the photos onto the mesh, so we have a fully textured 3D model as a result.
Estimated reading time: 2.5 minutes
Additional contents
sfm_export_model.jpg Chapter 7: Exporting the model for archiving
Lastly, we want to save our model in a way, that we can open it later. Tomorrow, next year and next century.
Estimated reading time: 1.5 minutes
c4d_final.jpg Chapter 8: Scaling the model
All is relative. So is your model so far. We have to scale it to the correct dimensions in order to really have it ready.
Estimated reading time: 3.5 minutes
screenshot_2021-11-10_153114.jpg Chapter 9: Combining two point clouds into one
If you want to create a real 3D model, you need to combine at least two different point clouds. Here, I explain how to do that by the example of a stone-age bi-face.
Estimated reading time: 5 minutes
  • sfm/start.txt
  • Last modified: 2024/02/11 19:50
  • by