I have two big point clouds (200m*200m) which i want to match. For these point clouds origin is (0, 0) and it extends only on the one side(i.e,. minimum and maximum points of the point cloud looks like [0, 0, 0] and [200, 200, 0]).
Whenever i try to match these point clouds, registration method provides me really good result near the origin until 40-50m from the center and then i can see a slight misalignment and as i go near the farthest point, then error in the matching is seen as 0.8-1m. So What i want to do is, considering the two point clouds have very less transformation between them (within 2m shift), to cut the point clouds in tiles (part by part, may be a part of size 40*40) and perform local matching and stitch them together again.
For doing this, I would have to shift the tiles to the center and perform matching there and align those point clouds and put it back where they were. But the problem seems to be simple but it gets harder when i implement it and it does not result in better result.
What would be ideal to perform such local matching away from the origin of the point cloud?
