LiDAR Depth-Map Sparse-to-Dense

asked 2020-01-30 10:14:44 -0500

Horst gravatar image

Hey Guys,

I want to convert a sparse (at the very bottom) depth-map to a dense one. I want to reach that by using bilateral filter (if that is successfull I want to apply Joint Bilateral etc.)..

But with the current information I'm quite confused. If I apply (and I have tried it) the bilateral filter to the raw depth-map, nothing happens. Why? Because it preserves the edges (as it is supposed to) and don't fill the unknown values (unknown values = zeros, because depth map is sparse and not dense) with interpolated ones. For that reason my approach was to look for x-nearest (kd-tree) known values and only apply the bilateral filter to that points, but even the search for that points took a lot of computational time.

What is the preprocessing step? Do I need to fill all unknown values manually? Is there any setting for the image filters which says "ignore all unknown/zero values"? Nobody exactly say what the do with the depth map before they apply the bilateral filter.

If someone has a hint, i would be very happy ! Thanks in advance


edit retag flag offensive close merge delete