README.md

Exploring 3D reconstruction

For a simple case using opencv's solvePnP algorithm.

To run the notebook locally

1. Create a Python virtual environment. For example to use conda to create an environment called 3d-reconstruction:

   conda create -n 3d-reconstruction python=3.10
   

2. Activate the virtual environment. With conda:

   conda activate 3d-reconstruction
   

3. Install the dependencies specified in the requirements.txt file with:

   pip install -r requirements.txt 
   

4. You should be able to run the compute_camera_extrinsics.ipynb notebook in the created virtual environment 🎉

Next steps

Multi-view triangulation:

• https://github.com/strawlab/pymvg/tree/master
• Gist to triangulate a point seen in n views
• This pycalib repo is educational but probably a good starting point.
  • Here they have the function to triangulate an array of points given their 2D coordinates in two or more views

Bundle adjustment:

• What is the difference with triangulation?
• This scipy tutorial seems useful.
• From the pycalib repo: this tutorial seems good.

Triangulate points -- opencv example

• only 2-by-2 views are possible in opencv's Python API
• see examples here

Some useful resources:

• OpenCV: Pose computation overview