ABSTRACT
Base Camera Pair ............................................................... 136 6.5 Camera Parameter Estimation Results ......................................................... 137
6.5.1 Camera Calibration Result................................................................ 137 6.5.2 Quantitative Evaluation in Simulated Environment ......................... 139
6.5.2.1 Free Camera Motion .......................................................... 140 6.5.2.2 Straight Camera Motion..................................................... 143
6.5.3 Qualitative Evaluation in Real Environment .................................... 145 6.6 Summary ...................................................................................................... 147 References .............................................................................................................. 148
In video see-through-based augmented reality (AR), estimating camera parameters is important for achieving geometric registration between real and virtual worlds. In general, extrinsic camera parameters (rotation and translation) are estimated by assuming xed intrinsic camera parameters (focal length, aspect ratio, principal points, and radial distortions). Early augmented reality applications assume the use of head-mounted displays (HMDs) for displaying augmented reality images to users. When using HMDs, changes in intrinsic camera parameters such as camera zooming are not used to prevent unnatural sensations in users; thus, assuming xed intrinsic camera parameters is not a problem in conventional augmented reality applications. In contrast, mobile augmented reality applications that use smartphones and tablet PCs have been widely developed in the recent years. In addition, augmented reality technology is often used in TV programs. In these applications, the changes in intrinsic camera parameter hardly give unnatural sensations. However, most of the augmented reality applications still assume xed intrinsic camera parameters. This is due to the difculty of estimating extrinsic and intrinsic camera parameters simultaneously. Removing the limitation of xed intrinsic camera parameters in camera parameter estimation opens possibilities in many augmented reality applications.
