Re: RGBD Outdoor Mapping - Offline Database Processing - Localization

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/RGBD-Outdoor-Mapping-Offline-Database-Processing-Localization-tp5258p5459.html

Hi Davide,

1. Depth accuracy for motion estimation. For loop closure detection, if you use scaled features like SURF/SIFT/ORB, the higher resolution is the better (you may have to tune their parameters on large resolutions), however more computation time is required.

2. If you are planing to do 2D mapping, you can use the middle ring of the lidar assuming the lidar is parallel to ground. For 3D mapping, use the full point cloud. With a low number of rings, you may increase the voxel filter size (0.1 to 0.5 meters). If you build rtabmap with libpointmatcher, ICP will be a lot better in particular with 3D point clouds. I would refer you to this paper, look the configuration used for the KITTI dataset (there are the ICP parameters used in different configurations of the paper). I am still working on a launch file for the VLP16, that could be useful for an example with that kind of lidar. Hope I will be able to publish it on rtabmap_ros repo in the next week or so.

3. GPS can be saved in the nodes of the graph. For visualization, they are used to align the graph accordingly to ENU coodinates and to export poses in GPS coordinates. They are used to reject bad loop closure detections. If Optimizer/PriorsIgnored is false (default true), they can be used during the graph optimization (this is however very experimental with g2o and GTSAM optimizer, as it is not always working, still an open issue).

cheers,
Mathieu