Clothooo / lvt2calib

A unified calibration for 3D LiDARs, visual cameras and thermal cameras
126 stars 13 forks source link

Doesn't find the calibration board on os1_32 #15

Open junji-code opened 3 months ago

junji-code commented 3 months ago

Hello, im running your code on the docker container provided and it doesn't find the calibration board. I've already tried changing the parameters with rqt_reconfigure but nothing happens anyways.

I have even filtered the data to cutout everything that is not the calib board but it doesnt work.

Can you help me looking if something on my data is wrong or if i'm doing something wrong as well?

Here is the link for my rosbag with my lidar data: https://drive.google.com/file/d/1OFpjm0hbMcirOwzK1v-oFvXuiEsHwzCb/view?usp=sharing

I also uploaded the raw data, without filtering the points. https://drive.google.com/file/d/1few2nxKRxKD6GLpGDpSSr7B6tggUZp18/view?usp=sharing

Clothooo commented 3 months ago

Hi, Junji. Thank you for your sharing. I noticed that the calibration board is too far from the LiDAR in your bag. Points on the board are too few, which makes it hard for our algorithm to detect the board. What's more, the scanning line of the LiDAR passing through each circular hole is only one, which does not meet the data standard for detecting circular holes. I suggest you control the distance between the calibration board and the LiDAR within 5m and ensure there are at least 2 scanning lines passing through each circular hole.

junji-code commented 3 months ago

I corrected what you pointed, and the algorithm successfully found the calibration board, but another issue happened when I tried to run the pattern-collection process:

OS1-32 image

OS2 image

Pattern-Collection window: image

Here is my rosbag with both point clouds: https://drive.google.com/file/d/1p_XJaV0e1pLBuI5hhoCQUmnHmgNOJ37x/view?usp=sharing

PS.: I ran the OS2 as a OS1_128 because of the point cloud density.

Clothooo commented 3 months ago

Hi Junji, the pattern-collection process froze due to the failure in circle detection. The feedback 'Not enough centers: 0' could be seen in your OS1-32 terminal and OS2 terminal. I checked your bag. Your input point cloud should reserve the background wall of the calibration board. which is crucial for circle detection. You can refer to this figure. Screenshot from 2024-03-28 14-03-46

junji-code commented 3 months ago

Okay, I missed that. So the pointclouds must be recorded at some indoor area? If so, will it work if I put some other plate behind the calibration plate for recording it on outdoor environment?