SphericalVoxelGrid Filter Contribution

[sddavis14]

I've been working on a project involving a LiDAR which has significant noise imposed on its returns. The angles are quite accurate, but the lengths of the returns are not. This noise does not lend itself well to a standard voxel grid because so much of the angular resolution of the point cloud must be sacrificed to accurately contain all of the points belonging to a given surface. I have implemented a new filter which downsamples the cloud using what I am calling a SphericalVoxelGrid. It takes a horizontal angular, a vertical angular, and a radial resolution as parameters and downsamples the cloud. Essentially, the filter is a voxel grid where the leaf size is specified in spherical coordinates instead of cartesian. This filter has delivered good results in my application because the radial resolution parameter can be set very low (long voxels) while maintaining a reasonable angular resolution.

Would this feature be a valuable contribution to PCL? If so, I will submit this feature as a pull request in the near future.

Thanks.

[SergioRAgostinho]

Given the context sounds like a worthy contribution. Some comments regarding the actual implementation:

• Your class will need to derive from pcl::Filter
• Check if there's anything from pcl::VoxelGrid you can reuse or even inherit
• Unit tests are mandatory and please write them in a way in which you can actually fully predict the outcome of your filter i.e., loading bunny.pcd or something similar and seeing that it works is cool but doesn't really tell us if everything is actually running well, so please try to devise a controllable synthetic case which manages to exploit everything you need to test.
• If possible write a small tutorial with images and showing exactly the problems you had originally and what drove you to implement this SphericalVoxelGrid filter and how it improved your results. It should make it easier for the users to understand the exact use cases in which this new filter shines.
• Try to structure your commits in a way which makes is easier for use to review: One for the class definition, other for test, other for the tutorial, etc... Things which logically make sense together. I advise reading about rebasing and interactive rebasing in git. It should help make your workflow more flexible.

That being said, good luck and once you submit the PR just wait for our feedback. We take some time to review things, it normally goes on for a couple of rounds but we always come back to revisit things :)

If you have any implementation questions at some point, just write them here.

[sddavis14]

Great. Thanks and expect a pull request within a month or so.

[stale[bot]]

Marking this as stale due to 30 days of inactivity. It will be closed in 7 days if no further activity occurs.

[themightyoarfish]

Has there been any progress on this? Sounds like a shame to have this useful feature linger.