Robust Fourier-Based Checkerboard Corner Detection for Camera Calibration

Precise localization of reference markers is crucial for the accuracy of target-based camera calibration. State-of-the art detectors, however, are sensitive to optical blur corrupting the image in many practical calibration scenarios. We propose a novel method for the sub-pixel refinement stage of common checkerboard target detectors. It uses the symmetry of checkerboard crossings and exploits the periodicity in the angular frequency domain when the origin of a polar coordinate system is centered at the crossing. The detector estimates the crossing center’s sub-pixel position by minimizing spurious frequency components that occur increasingly at ever larger distances from the crossing center.