This repository contains two implementations of finding skeletons from pointcloud data in Python.
This implementation is from this paper.
Workings:
- It divides the the the space containing the points into boxes and connects boxes which have points which seem to be part of the same object.
- Then it reduces the connections trying to maintain the general direction of the connection (the directionof the connection being in either x,y or z direction).
This algorithm is only implemented partially, here a link to the paper.
Workings:
- It selects a random set of points (we will call centers) from the total number of points and iteratively moves these points towards the medial center of the points in the neighborhood but adding a 'repulsion force' for other centers close by.
- Gradaully increasing the neighborhood size, this results in the centers lining up.
- Then it tries to connect (make branches of the skeleton) for the centers which line up nicely. The criteria for 'lining up nicely' is based on the eigenvectors of the set of points in the respective neighborhoods.
- It uses several complicated implementation procedures (here is their c++ implementation to connect different branches.
Step 3 and 4 are complicated to implement properly and are not well explained in the paper and therefor eventuelly I quit working on this. I had a try at this which can be seen by setting the variable try_make_skeleton to True
Both files L1-skeleton.py and skelTree.py will by default use the provided pointclouds in Data and should plot the outputs. Simply run python 'one of the files' and all should be good.
In both implementation one can adjust the number of centers or number of boxes used or import a different pointcloud.npy file at the start of the main. Search: "__main__": .
The following python packages are used:
- pptk
- matplotlib
- scipy
- pickle