Closed junji-code closed 2 months ago
Thanks for trying Discocal!. Discocal optimized not only the camera parameter but also the circle radius simultaneously. Therefore, the final circle radius may differ from the initial input, and we evaluate the calibration quality by the discrepancy between these two values. If the quality is low, you should check 1) detection results(image quality), 2) whether the input value(radius and distance) is correct or not, 3) the pinhole camera model assumption.
People frequently make mistakes with 2), so I recommend checking the original circle radius and distance again. If the input value is right, could you send me the images so that I can check what the problem is?
I've tried different distances and rotations of the pattern, but I dont see a clear way to take them to optimize the results. Even when theres a lot of scenes where the circles seem to be detected correctly, the results are not good. How can I improve it?