Hey @Sebastian_Egli 
I found this (a bit dated) document which has a very, very brief overview: Process Breakdown · OpenDroneMap/ODM Wiki · GitHub
In short, the steps are:
- Images metadata extraction (GPS location, coordinate system)
- Structure from motion (from images to
camera positions/orientationsandsparse point cloud). We use OpenSfM for this. - Multi view stereo (from camera + images to
dense point clouds) - Meshing (from points clouds to triangle meshes, we use mostly a Poisson Reconstruction approach but have support for 2.5D meshing as well)
- Texturing (from camera + images + meshes to
textured meshes) - Georeferencing (from local coordinates + GPS and/or ground control points to
real world coordinates) - DSM/DTM (from real world points to elevation models)
- Orthophoto generation (from textured meshes in real world coordinates to GeoTIFF)
If you are looking to contribute to the project, I’d recommend to focus on one single step at a time. It can be overwhelming to become proficient with all the steps all at once.
The multiband support for sequoia looks very interesting! I think this would be one of the first steps of the pipeline (right after Images metadata extraction ). We welcome contributions, so if you have other questions, feel free to ask!