Auto-extracting technique for checkerboard corners for endoscopy

In an endoscopy camera calibration, the serious distortion makes the corner detection more difficult, and requires more human intervention. To solve this problem, a flexible, simple and precise method to extract the checkerboard corners is proposed. This method uses a Gaussian filter to remove the image noise, and calculates the gradient feature twice, and roughly locates the corners by constraining the extreme values. Then, this method removes the noise points by checking the symmetry characteristics of the checkerboard corners. At last, the optimized location of corners is acquired. The experiment results show that this method is simple, and requires a small amount of calculation, and can extract all the corners without human intervention. The average re-projection error is less than 0.2 pixel.