Abstract

We consider the problem of camera pose estimation fora scenario where the camera may have continuous and un-known changes in its focal length. Understanding frame byframe changes in camera focal length is vital to accuratelyestimating camera pose and vital to accurately renderingvirtual objects in a scene with the correct perspective. How-ever, most approaches to camera calibration require geo-metric constraints from many frames or the observation ofa 3D calibration object — both of which may not be feasi-ble in augmented reality settings. This paper introduces acalibration object based on a flat lenticular array that cre-ates a color coded light-field whose observed color changesdepending on the angle from which it is viewed. We derivean approach to estimate the focal length of the camera andthe relative pose of an object from a single image. We characterize the performance of camera calibration across various focal lengths and camera models, and we demonstrate the advantages of the focal length estimation in rendering a virtual object in a video with constant zooming.