Point cloud processing acceleration solution

[cycyc1949]

What are the better acceleration solutions for point cloud processing?
Several suggestions that come to mind at the moment:

1. Downsampling: Downsampling is performed first and then motion compensation is performed
2. Roi filtering: Roi filtering is processed before downsampling, and the radar output is directly filtered once using R before converting to Cartesian coordinates

[knzo25]

@cycyc1949
We are implementing filtering in the driver in nebula as of now.
Downsampling may depend on the sensors used, and each integrator is free to add such a step before the other parts of the preprocessing.

If I can things going my way, I will add this pipeline as an option to the current one (almost the same but in CUDA, which makes thing "fast")
autowarefoundation/autoware.universe#9722