I have been trialing the R200 for machine vision/industrial inspection tasks, and every time I always find that the R200 depth data is terrible.
The data has huge "lumps" on surfaces that are flat, where the magnitude of these lumpy surfaces can reach over 10mm from peak to valley. This is from a distance of about 1 metre, and the target are various reasonably flat objects like plastic trays or wooden pallets etc.
Are there any techniques or hints/tips you guys can share to get much more accurate depth images, ones that do NOT have ridiculous lumpy surfaces?
To show you what i'm talking about, here is a depth cloud captured of a pallet:
Yeah, using a more expensive sensor, like a lidar.
If you have a sensor with a lot of noise like this one, you could try fitting models to the data, like a plane for example. Have a look at PCL for more details on filtering.