Improve rtabmap SLAM performance by enabling SURF feature

[simon.xm.lee]

Hi Mathieu,

My environment is  "KinectV2 + odometry". I use robot's odometry instead of rgbd_odometry. And the SURF feature seem just for visual_odometry(rgbd_odometry, stereo_rgbd_odometry). Therefore I think enabling SURF feature would not improve the SLAM performance at first.

However I see there are feature detection/extraction points showing in rtabmapviz. I guess the rtabmapviz would also include feature detection/extraction process. Am I right?

If yes, I would build opencv3.0 with extra module(xfeatures2d) to enable SURF feature.

Regards.

[matlabbe]

Hi Simon,

Features shown in loop closure detection view are those used for loop closure detection. They are independent of those used by visual odometry (which are GFTT/BRIEF by default). If SURF is not available, ORB features are used for loop closure detection. If built with SURF (xfeatures2d module of OpenCV3), SURF will be used by default for loop closure detection.

cheers,
Mathieu