search

koide3 / direct_visual_lidar_calibration

A toolbox for target-less LiDAR-camera calibration [ROS1/ROS2]
https://koide3.github.io/direct_visual_lidar_calibration/
636 stars 83 forks source link

Result is off and wrong projection during calibration #93

Open khuechuong opened 4 days ago

khuechuong commented 4 days ago

This is a follow up from issue 91. I did with my dataset with a pretty wide angle lens. So I did intrinsic calibration with kalibr. The preprocessing cloud point looks good. The problem was that at the end, it was still kind of off: result_rviz I did see that during the calibration process, there was some mis-projection in the corner that projects my april tags to the wrong: result

So I was wondering, how should I fix this? Should I go to an open environment with enough features like your rosbag?

koide3 commented 3 days ago

The viewer doesn't do hidden points culling, and thus this effect (back objects are colored by the pixels of front objects) itself doesn't mean a failure. Meanwhile, if you see misalignment on front objects, that means the calibration result is not satisfactory.

khuechuong commented 3 days ago

So then, how would I fix it?