what is the relation between visual features "loop closure" and navigation using Kinect

[m.omar82]

Dear Mathieu
I develop Ensemble loop closure detection using multiple feature descriptor based on rtabmap and I like to test it with rtabmap navigation. When I tracing rtabmap I can't find the relation between the visual features and the odometry data from kinect.

My questions:
1. if we improve the loop closure detection this will improve the navigation with kinect or stereo?
2. How can I extract the trajectory to calculator absolute trajectory error (ATE)?
3. can I use console Application to create map with  kinect or stereo dataset?

[matlabbe]

Hi,

1. yes, better localization is always good for navigation
2. do you mean in format used in TUM's RGB-D datasets? If so, the easiest way is to do File->Export poses (then select RGBD-SLAM format).
3. No, the console app is only used for loop closure detection benchmark. For RGBD-SLAM, you have to use the GUI. Note that you can disable 3D cloud rendering (Preferences->3D Rendering), close odometry view and loop closure view to avoid most of the GUI latency overhead.

cheers,
Mathieu