Issues with poses.txt generated by Kiss-ICP

[SvenMuc]

Hi, I'm trying to use the kiss-icp to generate a poses.txt file for the point_labeler tool. The kiss-icp poses_kitt.txt format seems to be correct, and can be read directly into the labeler tool. But the results are looking different.

Kiss-ICP Results The results are looking pretty good.

Point Labeler Results using the Kiss-ICP poses There seems to be mainly an issue with the heading.

Are there any differences in the coordinate frames or the interpretation of the poses format? If I use the default camera frame in the calib.txt, everything should be fine, isn't it?

[jbehley]

TLDR: Referring to your issue (https://github.com/PRBonn/kiss-icp/issues/167). When KISS does generate the poses in the LiDAR frame, then you have to modify the calib.txtsuch that the Tr is a identity matrix.

More details:

I pre/post multiply Tr in the loader such that it fits to the KITTI convention that all poses are given in the camera coordinates, therefore, one has to make P_lidar = inverse(Tr) P_cam Tr to get the pose in the LiDAR coordinates.

Thus, the easiest way is to modify the calib.txt such that Tr: 1 0 0 0 0 1 0 0 0 0 1 0.

[SvenMuc]

Thanks for your reply. Yes, that's what I exactly did and was also my expectation.

Maybe the root cause are the exported kiss-icp poses (see https://github.com/PRBonn/kiss-icp/issues/170). There seem to be a mismatch of the expected coordinate frames.

[jbehley]

could you solve the problem?

[SvenMuc]

We identified the root cause for the jumps. I occurs everything we faced some frame drops. In that case the matching fails due to the bigger distance.

[jbehley]

Thanks for the info and letting us know. Thus, it seems like the issue is solved and can be closed?

[SvenMuc]

yes, from side it is solved. Thank you for your support.