Model-based Pose Estimation in an Urban Environment

This paper addresses pose estimation problem for augmented reality applications in an urban environment. The pose is estimated based on the known wire-frame models of buildings and line segments detected on the images. One of key problems for model-based pose estimation is to establish the correspondence between the model and detected features. The main contribution of this paper is a robust matching method that combines geometric and intensity constraints to establish accurate correspondence for pose estimation. Our pose estimation method is based on the EKF framework and the experiments show that the proposed method can estimated camera pose reliably even for abrupt and large inter motions.