Point Cloud Reconstruction from Stereo Images

After cloning the git repository run the following commands in the project root folder

python3.11 -m venv venv
source venv/bin/activate
pip install -r requirements.txt

Now you can run the implemented stereo algorithms with

python main.py --method sgm --data images/cones/

you can append the --subpixel flag to use subpixel interpolation. The --method argument accepts the following values [bm, sgm, opencv]. At the beginning of the main.py you can find a full list of arguments. A PNG of the disparity map and the .ply file of the point cloud are stored in the output folder.

Before testing the implementation of blockmatching in C you first have to create the shared objects from the C files in the c folder. You can do this by running

invoke build-block-matching

or for building the C algorithm with OpenMP run

invoke build-omp-block-matching

Both compilation steps have been tested on MacOS 13.4.1 and Ubuntu 22.04. If you want to run the project on windows or change the path to OpenMP you have to modify the command in task.py.

After creating the shared objects (.so files) you can run

python main.py --method bm --language c

To visualize the generated point cloud run

python render.py --file output/<YOUR_PLY_FILE>