Re: RGBD SLAM and then RGB Localization

Posted by matlabbe on
URL: http://official-rtab-map-forum.206.s1.nabble.com/RGBD-SLAM-and-then-RGB-Localization-tp489p524.html

I think you could test the RGB-only localization part using the same camera (Xtion without the depth) to see if you can have more inliers.

Also, you could have more inliers if you re-extract the features between matched images. By default, rtabmap extracts a maximum of 400 SURF features from each image when creating the map. So if your image you are trying to localize is not directly in the same field of view of the best matched image of the map, you may start with low matches. By re-extracting the features, you could extract more like 1000 features or even no limit. I have done this kind of process when evaluating loop closures (which don't have generally not many matches on large-scale). Code is here. "dataTo" would be the best match image on the map (RGB-D image) and "dataFrom" the new image (RGB-only)... it will only work if Memory::computeVisualTransform() uses PnP (for which the depth is not required for the second image).

cheers