Robust focal length estimation based on minimal solution method

Abstract

We present a new approach to estimate the focal length for camera calibration in multiview reconstruction. As a popular camera calibration approach, minimal solution method gives rise to a great number of focal-length estimates, from which generating an accurate one is of great significance. Our method concentrates on how to obtain an accurate estimate and is carried out in two steps: firstly, a norm constraint for the fundamental matrix is employed to prune the low-confidence focal-length candidates. Then the focal-length estimate is obtained with a robust focal-length estimation scheme, which consists of occurrence to probability transform, focal-length candidates resample and final estimation with expectation. Experimental results demonstrate that our method could obtain better estimate with both higher accuracy and higher stability than the state-of-the-art method.