Abstract
This paper proposes a method for estimating extrinsic camera parameters using video images and sparsely acquired GPS data. Although the extended bundle adjustment methods which fuse vision and GPS data have been already proposed and they work finely if GPS data are accurately acquired, existing methods have the following problems: (1) The accuracy of estimated camera position largely depends on the confidence of GPS positioning data because they do not consider the GPS positioning confidence.
(2) Solution often converges to local minima when GPS data can not be acquired for long time. In order to solve these problems, we add weighting coefficients depending on the GPS positioning confidence to an energy function for extended bundle adjustment. In order to avoid local minima, estimated camera positions without GPS data are fitted to GPS positions before optimization.
