search

shichaoy / semantic_3d_mapping

Semantic 3D Occupancy Mapping through Efficient High Order CRFs
Other
192 stars 60 forks source link

Does ORB SLAM and depth estimation use the gray camera? #17

Open zeroAska opened 5 years ago

zeroAska commented 5 years ago

Hi, when we trying to run kitti15, our 3D scrolling map seemed to be drifting. We are not sure if we generate the depth images and the orb slam poses. Currently, we generate them using kitti's gray camera (and its calibration matrix). When you generate kitti05's depth images and poses, did you convert the rgb left camera into gray, or you directly use the gray camera?

Thanks!

shichaoy commented 5 years ago

@zeroAska First remember to change calibration in main_stereo_images.cpp I use gray camera for the orb slam pose estimation (their provide calibration file). For elas depth estimation, I read color image and convert to gray. For the online grid mapping, I un-project RGB image to 3D grid. For your problems, if the scrolling map is drifting, that seems more related to orb slam pose estimation, can you first verify their result is reasonable? maybe you use wrong calibration for orb slam.

SrOdranoel commented 3 weeks ago

hello. Were you able to calibrate using Kalibr or did you use any Matlab or OpenCV libraries? I looked at the ORB-SLam 3 documentation and they recommend using Kalibr but that standard kalibr on A0 sheet is no longer available. I'm thinking about calibrating using the math lab library created by Jean-Yves Bougue (Camera Calibration Toolbox for Matlab).