An Accurate Corner Detection Algorithm in the Infrared Image

In HCI (Human-Computer Interaction), the camera is often with wide angle lens and the captured image is usually distorted. The infrared image is more blurred than general image, so the contrast and the brightness is lower. In this paper, a cross shape mask is used to get the preliminary chessboard internal corners. And the adaptive threshold is put forward to remove outliers and align the corners. With considering the local symmetry property around the internal corner and the chessboard line architecture, the final internal corner coordinates with the subpixel accuracy can be computed with the Orthogonal Vector. Some traditional algorithms are experimented to show its improvement.