Tool to real-time / fast cluster well-separated point clouds on python

Question

I am looking for a library, preferably in python, to cluster a point cloud, a little noisy but well separated like this image in (near) real-time.

I have tried with HDBSCAN library, with boruvka_kdtree algorithm, but can get around 3s for 30k-point point clouds. The benchmark says this could be the fastest algorithm already. Here are a few points, in case they could help: each cluster should be at least .5x.5x.5 meters, and should contains at least 100 points.

If you could possible suggest another library, or a different implementation of DBSCAN, or a good set of argument for HDBSCAN to run this problem in real time. A little sacrifices of accuracy could be possible as long as it's real time speed.

Answer 1

On such data, a grid based implementation will shine. No need to use HDBSCAN* - regular DBSCAN or an even simpler grid cell merging approach will be faster.

You just need to write them yourself there is no library.