Robotics StackExchange | Archived questions

Accounting for distortion in ar_track_alvar

Hello,

I am trying to use artrackalvar to get pose estimation of april tags. As noted here, there is an issue with the package when it comes for accounting for distortion!

Here we can see that the issue remains, as the Distort and UnDistort classes are still commented out. I am also noticing some systematic errors in the poses that seem consistent with the hypothesis that image rectification is not correctly taken into account.

Has anyone found a way to hack around this?

I have tried to naively rectify the image beforehand, as well as to simply uncomment the Distort and Undistort classes in the artrackalvar code, but neither seemed to work well.

Thanks!

Asked by EugVal on 2019-12-12 12:02:54 UTC

Comments

I'm not sure, and I'm not really into computer vision topics, but isn't the idea that you feed ar_track_alvar already undistorted images? Nodes/nodelets in image_proc were created for this purpose.

Asked by gvdhoorn on 2019-12-14 02:01:42 UTC

The documentation originally suggests that you have to subscribe to the un-rectified image topic : camera_image (string) -- The name of the topic that provides camera frames for detecting the AR tags. This can be mono or color, but should be an UNrectified image, since rectification takes place in this package . That said, I did try what you suggest with image_proc, but it didn't seem to get me better results either so I'm still trying to figure this out!

Asked by EugVal on 2019-12-14 05:54:54 UTC

So just making sure: you configured everything to use the rectified image and used the node(lets) in image_proc to rectify the image, correct?

Asked by gvdhoorn on 2019-12-14 11:32:07 UTC

I used image_proc to rectify the image, then input the rectified image toar_track_alvar. I didn't change any config on ar_track_alvar, as I don't see any arguments that allow me to specify whether I want to use a rectified image or not. :/

Asked by EugVal on 2019-12-17 04:00:20 UTC

Answers