Using different features for loop closure and transformation estimation

[tango_explorer]

Hi,

I'm fiddling with the idea of using different features for loop closure detection and transformation estimation. Right now, I'm thinking that the best way to go about this would be to add a separate keypoint/descriptor data object to sensor data class and work with that. I was wondering if anyone knows of an easier way to do this and what files would I need to change to make sure I don't break anything dependent on having the features the same for loop closure and transformation estimation?

Thanks

[matlabbe]

Hi,

It is already possible to choose different features for loop closure detection and transformation estimation. Loop closure detection feature detector is defined by parameter "Kp/DetectorStrategy" (default SURF). BOW will be done on this type of feature. When a loop closure is detected, we can reextract other features for transformation estimation, see parameter "RGBD/LoopClosureReextractFeatures". When this parameter is enabled, parameter "Vis/FeatureType" (default GFTT/BRIEF) is used to extract features for transformation estimation.

cheers,
Mathieu