How Many Images for Camera Calibration?
I am wondering how many images are recommended to calibrate a camera using the chessboard corners approach. Should I use 5, 10, 20? Also, if 10 is a good number, for example, is there any benefit of using more?
Further, is there any advantage to moving the chessboard around in the image? That is, should I try and take pictures with the chessboard placed at all the parts of the image or are multiple images of the chessboard at the same location just as good.
Any help appreciated.
Regards, James
"is there any advantage to moving the chessboard around in the image" - definitely. just think about, what 'barrel distortion' means.
"if 10 is a good number, for example, is there any benefit of using more?" please check the return value of calibrateCamera(). if it's going down you're doing better, else you probably overfit.
Have a look at these blogs, which is really useful for stereo calibration.
http://blog.martinperis.com/2011/08/s...
http://blog.martinperis.com/2011/01/o...
Number of images and orientation of chess board being captured matters a lot. It is recommended to capture as many as 30 to 40 images in different orientation and different distance to make calibration best.
Thanks Berak and Spark. There is a significant processing time difference between 20 and 40, e.g., HD images and I was trying to get a sense of whether it makes sense to incur that. All of the examples I have seen first collect a bunch of points from images before calling calibrateCamera to get a result. Thus, you need to know the number of images first.
With regard to moving the chessboard around, what I was thinking is that because the output of the calibration is not mask for each pixel but coefficients of a function representing the distortion, coverage might not really matter as long as you have a base idea of what the distortion is. However, it makes sense that more you move the chessboard around, the greater refinement you might get in the coefficients.