OpenCV cannot find chessboard corners in fish-eye images

Question

I am using OpenCV3.0 and VS2013 to do fisheye wide-angle(185 deg) camera calibration. There are some pictures that can work well, e.g:

and the correction result is satisfying as:

![corrected image -- fish-eye][2]

The result looks good, but I use a similar image to do calibration instead, it cannot find any chessboard corners!! The image is like this:

The results are very confusing to me since I try many many similar images( same distance different angel, same angel different distance ) for about 60 images, and small part of them can be detected. I use 2592x1944 resolution BMP files, and I accurately put the chessboard center in the center of camera, but the result is no good than the casually captured images, some even cannot find corners. I use this code to detect:

bool patternfound = findChessboardCorners(imageGray, board_size, corners,CALIB_CB_ADAPTIVE_THRESH + CALIB_CB_NORMALIZE_IMAGE+ 
        CALIB_CB_FAST_CHECK );

Answer 1

Turn off CALIB_CB_FAST_CHECK , when testing fisheye image.