Closed bhushanap closed 1 year ago
I think the board is moved in the same plane, so you likely need to "pitch" the whole whiteboard forward and backwards. Otherwise the dataset looks ok besides not exciting this board in all 6dof motion. Additionally, it is only in the middle of the image, but you should ensure to move it to all regions of the FOV. Either you can move the camera and keep the board static, or move the board in full 6dof motion (full rotation, scale, translation, and skew in the FOV of the camera).
The full FOV is particularly important as detections near the edge of the FOV are key to accurate distortion estimation.
I am really struggling to get low distortion values for even a single camera calibration. I wanted to run stereo cameras but for some reason the values are extremely weird.
Earlier I was trying both cameras at once, and that failed. So I started to do mono camera to see if that was the issue. And indeed each of the camera projections is bad. I am using a standard April 6x6 but a slightly changed yaml file because of some printing issue. I have tried multiple both equi and radtan but to no avail. I tried initializing the focus values to multiple values now. I have also tried making multiple bag files, mono, color, raw, compressed but none of them seem to solve the distortions issue.
Here's the log of the output I get for various initializations of focal length for pinhole-equi on raw camera data:
The cameras and lenses used are for pointgray cameras.
This is the static bag file I have been using for calibration. As you can see, the tags are detected well.
Any ideas for how I can improve calibration?