Camera Calibration Error Modeling and Its Impact on Visual Positioning

Abstract

Unmodelled calibration errors can lead to integrity risks in visual positioning. In this paper, we present a method to estimate the uncertainty of camera intrinsic parameters induced during the calibration process by backpropagation of the uncertainty in the 3D reconstruction error. The conventional approach propagates the uncertainty of image measurements forward and suffers from its high nonlinearity and the mismatch between the real and the applied camera model. Our proposed method is completely independent of the estimation techniques used during the calibration phase and can therefore be used to evaluate the performance of the calibration. Additionally, we also demonstrate how to propagate the uncertainties of calibration and image measurements into the estimated camera position. The derived camera position distribution shows a high concordance with Monte-Carlo results.