Statistical lower bound for variance of checkerboard pose estimate

Checkerboards (and fiducial markers) can be used to estimate relative camera pose. Knowing the accuracy of that estimate is important for many applications, for example, sensor fusion. Thus, this paper presents an analytic model of the Cramer-Rao lower bound for checkerboard pose estimation, which gives the variance of the minimum-variance unbiased estimator. Both Monte Carlo simulations and real data are used to verify the model. The model is parameterised by pose, checkerboard configuration and camera intrinsics, so it can be used to predict the accuracy of an estimator before data is available as well as to evaluate the accuracy of a specific estimator implementation. This approach for predicting and evaluating checkerboard pose estimation accuracy has not been considered in prior research.