Open dagata-mining opened 2 years ago
Always happy to get new filters!
I can see advantages for extracting local visible maps in a large point cloud when doing real-time localization. Do you have screenshots of before and after? I'm curious about the water effect. But we ofter observe reflection caused by glasses, but there is typically a hole on the glass so it's hard to confirm that we are behind a surface :
Yes, here's an example of a scan where the points where going through a mesh grid with small holes. The points returned from the mesh where blurry : Here's the texture:
Here's the before (yellow points are the unwanted points):
Here's the after with a 2.5cm octree size and a fparam of 2
You can send me some samples with the origin view point origin at (0,0,0) and I'll run it to see the results.
The data can be retrieved from here: https://projects.asl.ethz.ch/datasets/doku.php?id=laserregistration:apartment:home
I think pose id 6-7 see the windows.
Hi François,
From working with lidar scans in various environments (Mostly underground mines), I've come to realize that water holes, grid meshes or glass often tend to create unprecise points behind the surface. This affects the Icp and induce an unwanted error. So I implemented on my side a filter based on a hidden point removal approach: https://www.researchgate.net/publication/200018717_Direct_visibility_of_point_sets I implemented an octree based approach for fast hidden point removal in my SLAM. It worked pretty well on my side. I would see a that approach a new type of filter for Libpointmatcher. As you can see in my bit of code I'm using PCL, I'm sure there is way to implement it without it, and I'd like to investigate that, but I wanted to seek out the interest of this approach on your guys side, or if it has already been implemented.