search

koide3 / direct_visual_lidar_calibration

A toolbox for target-less LiDAR-camera calibration [ROS1/ROS2]
https://koide3.github.io/direct_visual_lidar_calibration/
702 stars 102 forks source link

Understanding results #75

Open jiheddachraoui opened 5 months ago

jiheddachraoui commented 5 months ago

Hi,

I'm not sure how to tell if the solution and calibration are working well. The cost function is 0.98, but I'm not sure what that means. Also, I couldn't understand the results from the viewer. Since I'm using the superglue method, I'm not sure if I should trust the initial guesses.

Also, does the lidar and camera have to be mounted exactly like in your video? In my setup, the camera is on both the bottom and the side of the lidar.

Thank you.

koide3 commented 5 months ago

Assessment of calibration results is actually a deep topic as there is not groundtruth. In my opinion, in practice, we can do only check if images are well mapped on the point clouds with our eyes as in this video https://www.youtube.com/watch?v=yTz07B23NXM&t=3s.

This package does not assume anything about the LiDAR-camera transformation. You can place the camera anywhere as long as it is rigidly attached to the LiDAR.