Scan drift on my first scanning project 1965 airstream

[Jonas Thomas]

Hello, I just started using Rtab-map over the weekend and I still have huge amount to learn. My wife (and I) are gutting a 1965 airstream to restore it. We are down to the bare walls and I added blue tape markers for the software to pick up some references when scanning. It seems that if I go back to fill in a hole, I get this weird drift. I'm not sure at this point what I'm doing wrong. (I'm using a kinnect v1 for the scanner) It seems that some sort of error is being created and I get a new wall instead of appending to the image it seems to create a new one. I tried running the post process and it doesn't seem to fix anything. I don't know if it's my scanning technique/ a limitation of what the Kinect can do. In order to maintain marital bliss, I'm a bit under the gun here. My wife is highly modified to get the airstream down to the studs. Any tips one what I'm doing wrong would be appreciated.

[Jonas Thomas]

I was looking this link https://www.youtube.com/watch?v=_qiLAWp7AqQ

It seems that some of my issues may be settings related to  loop closure detection?

At the suggestion of some of my friends at the makerspace I belong to I tried a few things that seems to have improved things a pit.   Since the interior is smooth and uniformly colored I added some blue tape to the interior.    I also mounted the kinnect on a tripod and did a 360 scan which seemed to improved things immensely,  but that means I need to stitch the scans together in meshlab.

This is very interesting software.  Thank you for creating it.

[matlabbe]

Double walls effect is caused by pose estimation drift. Most drift can be corrected by loop closure, though being not always perfect. Visual features are required for good tracking (less drift) and to find those loop closures. Make sure the lighting is good, don't scan too close from surfaces so that the camera can track more features. See also this tutorial: https://github.com/introlab/rtabmap/wiki/Kinect-mapping