Velodyne 16 line radar cannot complete calibration about direct_visual_lidar_calibration

[Ning030]

This is a ROS1 bag that I recorded myself.
https://drive.google.com/file/d/1Z_S8l-WU6jQWuUJfvNQVFbGRWMgjAY2d/view?usp=sharing
This is a robosence 16 line LiDAR .
Unable to generate dense point clouds during the data processing stage!!!

[Ning030]

Have you solved this problem? If so, can you provide a solution? Thanks Thanks!

[koide3]

Can anybody provide an example rosbag to reproduce the issue?

[af-doom]

Have you solved this problem? If so, can you provide a solution? Thanks Thanks!

sorry ,no

[koide3]

Hi @Ning030 ,
Thanks for providing the rosbag. I found that the issue stems from input point clouds containing NaN points. I added a filter to remove invalid points and pushed the updated code to vlp16 branch, which will be merged into main soon.

[Ning030]

Thank you for your answer. I greatly admire your excellent work！！
I would like to ask, how to use the automatic initial guess（SuperGlue）?
I downloaded the source code, but the calibration effect is not good.
Did you retrain it？

[koide3]

The accumulation result may not have a sufficient point density required for the automatic initial guess pipeline. Try to increase the LiDAR data (with up-down movement) or to use the manual initial guess interface.

[Ning030]

Thank you for your suggestion.
But I still have poor calibration results when using the data you provided to run the algorithm. （With automatic initial guess）.
Why is this?

[koide3]

Which data? and how the data was poor?

[Ning030]

Sorry！
I found out that it was caused by my input of wrong camera intrinsic data.
Thank you for making this work public.