Camera Calibration Based on Phase Estimation

Abstract

We propose a novel camera calibration method based on phase estimation of a periodic grid pattern that is more accurate and robust to image blur, noise, and uneven illumination. Instead of intensity-based geometric feature detection from the calibration image, we extract the phase information to acquire 3-D-to-2-D point correspondence. The calibration procedure includes three steps: 1) estimate the tilt angles of input images based on a pyramid geometry model and remap the images to a canonical fronto-parallel plane to correct perspective distortion; 2) use a highly accurate phase estimation method to localize control points; and 3) apply an iterative refinement method to compute the camera parameters until convergence. Using the perspective correction and phase estimation method together help increase the accuracy of control point localization and consequently of camera calibration. Synthetic as well as real images from high accuracy patterns etched on ceramic plates and low accuracy patterns printed on article show that the proposed method outperforms the most widely-used methods, in terms of accuracy and robustness.