Point Cloud SLAM without RGB Information [closed]

asked 2012-06-19 01:23:28 -0500

ChickenSoup
387 ●28 ●35 ●36

updated 2016-05-17 02:45:14 -0500

Hi all,

As far as I understand, RGBD SLAM uses visual features in addition to the point cloud to perform SLAM. If I use a tilting LRF, which has high resolution and range than a kinect, can I use RGBD SLAM package (probably with some modification) to do just point cloud SLAM without using any visual features?

Thanks in advance

Closed for the following reason the question is answered, right answer was accepted by Martin Günther
close date 2016-05-17 03:25:54.488106

answered 2012-06-19 02:57:25 -0500

Felix Endres

6468 ●28 ●57 ●93 http://www.informatik....

Yes and No. You could disable Features by sending a blank depth image and enable GICP to estimate the motion, but GICP is not used for finding large loop closures, since the convergence of ICP does not necessary mean you are matching the same scene. This problem is of course less pronounced with the FOV of a LRF.

A good way to go would be to extract 3D features (NARF, Spin Images) and match these. But RGBDSLAM doesn't to this, so you would need to modify it on the source code level.

edit flag offensive delete link

Comments

Thanks @Felix Endres. I will go through the RGBDSLAM code and try to modify as you suggested.

ChickenSoup ( 2012-06-19 06:11:36 -0500 )edit

If you want to replace the visual features with 3d features, the easiest way should be to add a new callback to openni_listener for a point_cloud (or depth image and camera_info) and then modify the Node class to your needs.