AlejandroSilvestri

Results 68 comments of AlejandroSilvestri

ORB-SLAM2 outperforms ORB-SLAM, without drawbacks. Monocular visual SLAM never gets real scale. You must convert from monocular slam arbitrary reference system to some real world reference system of your choice....

Well, monocular SLAM doesn't get the right scale. Each time ORB-SLAM and ORB-SLAM2 initialize, they pick an arbitrary scale based on the ser of map points they were able to...

Let me know how you do that when you succeed!

@ThibaudMZN , I'm glad to hear about this. I myself am working on something like that, but with a different approach. I assume you want to apply a 7dof transform...

Hi @RameshKamath, I share some thoughts. The only influence of scale in a euclidean transformation matrix (a rototranslation matrix) is the magnitude of the translation vector submatrix (the right column...

@RameshKamath I think the best place to change world scale is in global bundle adjustment, or directly right during initialization. All subsequent frames will adopt this scale. [ORB-SLAM3](https://github.com/UZ-SLAMLab/ORB_SLAM3), still in...

@RameshKamath You need to apply new poses to keyframes too. If that still doesn't work, you can always apply a conversion to the output current frame pose: let orb-slam2 use...

@RameshKamath You should get only one scale factor from translation, not three. You are calculating three independent scale factors for each axis, this stretches your space. If orb-slam2 and imu...