Optic Illusion Patterns for Accurate Camera Calibration

Gaku Nakano, Takenobu Kiyama

Abstract: This paper proposes a camera calibration pattern that leverages optical illusions. An optical illusion pattern is a texture that appears geometrically distorted but looks flat when viewed from the camera's perspective. The proposed method initially displays a conventional checkerboard pattern on a monitor to estimate the homography matrix between the pattern and the camera, along with lens distortion. Subsequently, the inverse homography transformation is applied to the pattern image to display an optical illusion pattern, and corner points on the illusion pattern are detected. As the captured image of the illusion pattern appears front-facing, the localization accuracy of the corner points is enhanced. The proposed method addresses the “chicken-and-egg” problem of needing to tilt the pattern for stable vanishing point estimation while equalizing localization accuracy of corner points at different depths. Simulations and real image experiments report that the proposed method outperforms conventional calibration methods for both narrow-angle and wide-angle lenses.

External IDs:dblp:conf/iccv/NakanoK25