Calibration of a moving zoom-lens camera for Augmented Reality applications

In Augmented Reality systems, real time integrating 3D virtual objects into the real environment requires camera calibration techniques for spatial alignement. The accuracy of the parameters is then a crutial task for designing an effective Augmented Reality (AR) systems, and the challenge to meet this requirement is made harder when dealing with moving zoom-lens cameras. Although recent approaches have been proposed using 2D/3D patterns, the use of zooming lens cameras brings new problems for robust and continuous estimation of the intrinsic parameters. This paper propose a robust algorithm for variable focal length calibration combining interferometry and 2D pattern detection to tackle the problem. A global optimization routine is driven to search for the best calibration parameters that fit the system. After a description of the different methods involved in each step of the process, the approach is validated with AR experimental results on a real time AR touristic application.

I. Introduction

Augmented Reality applications supplements reality with virtual objects to enhance users perception of the environment. This requires accurate calibration parameters of the system to spatially insert 3D virtual objects into the real environment. While extrinsic camera parameters should be estimated in real time for each frame, intrinsic ones are generally estimated during an offline preprocessing step and keep unchanged over time. For video-based systems, an extensive body of literature exists on camera calibration techniques [6], [5], based on 2D/3D pattern analysis. However, [3] showed that these intrinsic parameters, considering a pinhole camera model, are not independent subject to zooming and hence making the latter technique of [6] not suited to address this issue. Pan-tilt-zoom camera calibrations methods were studied and exposed in [7], [8]. Using standard calibration methods reveal in this case imprecision and discontinuity of these parameters. We propose in this paper a new approach for estimating the intrinsic parameters of the pinhole camera model by simplifying the calibration pattern process. The proposed process involves both interferometry and image processing techniques and a global optimization routine is driven to search for the best calibration parameters that fit the system making them more effective. In this paper, after over-viewing the adopted approach, we present some experimental results applied to a “telescope” deployed for a touristic AR Application (Fig. 1). Fig. 1. Telescope prototype located on the “Arc de Triomphe” in Paris