Change feature detection method in rtabmap_ros

[dmkarage]

Hi,

I am really new to ROS and Rtabmap as well. I want to implement Rtabmap in a differential wheeled robot which carries an Xtion Pro sensor. I would like to test different feature detection methods. I have managed to do it in the stand alone  application but I want to do it also through the ROS package. I changed the following parameters:

 
 

in the rtabmap.launch file. However I do not see any difference (visually at least) in the odometry screen of the rtabmapviz GUI. Should I also change some other parameters? Also which feature detection method is the most appropriate or on which factors does it depend?

Sorry for the long post and the beginner's level question.

Thank you  

[matlabbe]

We cannot see what you changed. Click on More -> Raw text to copy paste launch files.

If you are using rtabmap.launch, you can change features like this:

$ roslaunch rtabmap_ros rtabmap.launch rtabmap_args:="--delete_db_on_start --Vis/FeatureType 5 --Kp/DetectorStrategy 1"

Here using GFTT/FREAK for odometry and SIFT for the vocabulary of loop closure detection.

Binary descriptors (ORB, BRIEF, FREAK) are more suitable for odometry, for which we want very fast extraction/matching. Float descriptors (SIFT, SURF) are more suitablefor loop closure detection, because they are more discriminative (at higher computation cost).

cheers