Distortion in overlaps of database on RTAB-Map for iOS

classic Classic list List threaded Threaded
3 messages Options
Reply | Threaded
Open this post in threaded view
|

Distortion in overlaps of database on RTAB-Map for iOS

Daniel_Merizalde
Hello,

I am trying to create a 3D model of a tunnel using RTAB-Map on an Iphone 12 pro and getting good results, but because the iphone travels trough length of the tunnel twice, the data is captured twice and some times that causes overlapping and distortion of the 3D model as seen in the screenshot below:



Is there any way to set a timer so the capture dos not start immediatly? or ignore the fist part of the database? which is the one causing the issues. Is there some other way I could handle this issue? Like changing parameters like loop closure threshold and similarity treshold.


Thank you for your help!

Daniel Merizalde R.
Reply | Threaded
Open this post in threaded view
|

Re: Distortion in overlaps of database on RTAB-Map for iOS

matlabbe
Administrator
It is difficult to see, but ideally would be to detect loop closures when coming back.

To export only one part of the map, you can open Visibility panel in rtabmap, then uncheck clouds you don't want to export.

Another way is to open the database with rtabmap-databaseViewer, open Constraints View, scroll Neighbor links to last or first node you want to start assembling the clouds, then click on Reject button to split the graph. After that, in GraphView, you can set the root id to one node you want to include, then the graph will be optimized from there. When viewing or exporting the clouds, only nodes in the graph shown in Graph View will be exported.

cheers,
Mathieu
Reply | Threaded
Open this post in threaded view
|

Re: Distortion in overlaps of database on RTAB-Map for iOS

Daniel_Merizalde
Hello Mathiew,

I tried the options you described and they both seem to be working well. I will still try to adjust the speed and lighting to see if it can detect the loop closures but otherwise I believe the issue is solved.

Thank you very much for your help!!

Daniel Merizalde R.