Quality assessment of geometric camera calibration

Geometric camera calibration is a prerequisite to numerous methods of optical metrology and computer vision. We argue that typical calibration workflows fail to adequately address the quality of resulting parameters. Common reporting standards for the calibration outcomes are similarly inconsistent and reduce the transparency and reproducibility of camera-based measurements. We suggest a novel way to quantify the consistency, reproducibility, and reliability of calibration. The method employs the well-known active target technique, extended in order to account for the uncertainties in the positions of calibration points. This approach is compatible with any camera models and may be integrated into existing calibration tools. Moreover, it may lead to novel algorithms to estimate true posterior distributions over model parameters while integrating the available prior knowledge. Such approach would be a sound solution for demanding metrological tasks. The method is demonstrated with synthetic (rendering-based) and real experiments (with a high-quality consumer camera), where we compare the algorithm of the OpenCV library with the recently suggested smooth generic camera calibration.