search

ankitdhall / lidar_camera_calibration

ROS package to find a rigid-body transformation between a LiDAR and a camera for "LiDAR-Camera Calibration using 3D-3D Point correspondences"
http://arxiv.org/abs/1705.09785
GNU General Public License v3.0
1.49k stars 460 forks source link

Calibration is wrong #20

Closed emilaz closed 6 years ago

emilaz commented 6 years ago

Hello, the results of the provided calibration tool seem to be very off. I tried running it a couple of times but the end result is often as can be seen below:

faulty result

To reproduce this result, I am uploading an appropriate bag file and all the needed config files: test.bag.zip conf.zip find_transform.launch.zip zed_left_uurmi.ini.zip

and the corresponding results I got: results3.txt

I am thankful for any suggestion on how to improve the result.

karnikram commented 6 years ago

Hey,

I haven't looked into your data completely, but it seems like your markers are positioned incorrectly.

The smaller valued marker should always be on the left!

emilaz commented 6 years ago

Hey, Marker #904 is on the left, #905 on the right.

karnikram commented 6 years ago

My bad. 905 looked like 903 on my system.

Anyway, I ran your data and I seem to be getting the right result.

Transformation , T :

T = [0.996554 -0.00467748 -0.0828156 0.0894906 0.0109373 0.997101 0.0752965 0.156528 0.0822233 -0.0759428 0.993716 0.388588 0 0 0 1];

userresult

emilaz commented 6 years ago

So yours is the average transformation after 100 iterations? I don't understand what the final rotation is telling me, since its values always seem to differ greatly from the average rotation. Also, would you be able to provide the visualization tool you used for the above graphic?

Greets, Emil

karnikram commented 6 years ago

Yes, I used the average transformation.

The script can be found here - #19

karnikram commented 6 years ago

Sorry, missed your other question. final_rot = average_rot * initial_rot

initial_rot is usually (pi/2,-pi/2,0) depending on your configuration

SinghiDivyanshu commented 4 years ago

hi @karnikram @emilaz , i,m stuck after getting the rigid body transformation matrix, now don't know how to visualize the output (point clound and image in same frame). here i'm using the monocular camera.