Robotics StackExchange | Archived questions

Best package to estimate object pose using camera

How would you recommend one use vision to determine the pose of an object for which there is a CAD model (and mesh)? One option is visp, but I think that only works for a 2D camera and I don't want to have to depend on camera calibration. Another option, I would think, would be to use a depth camera and determine pose by somehow comparing the model to the point cloud. (This depends on camera calibration, too, but it would be done at the factory.) Do you know of any methods, preferably ROS-based but not necessarily, for getting pose from a point cloud?

tl;dr --- How do you suggest getting pose from a camera? visp, depth camera, or other?

Asked by raequin on 2018-05-24 13:23:07 UTC

Comments

It's not open-source, but Halcon can do 6d pose estimation from cad+2d image.

Asked by gvdhoorn on 2018-05-28 11:05:47 UTC

To add: "6D pose estimation" is the keyword to find loads of very recent research papers on this, many freely available on arXiv. I don't know if ROS packages exist, but you could try wrapping the source code of any papers that make it available.

Asked by fvd on 2018-09-05 02:34:00 UTC

maybe this is relevant: https://paperswithcode.com/task/6d-pose-estimation

Asked by fbelmonteklein on 2019-05-06 09:10:52 UTC

Answers