| Augmented Reality (AR) is an interface technology designed to increase the efficiency of user's interaction with the real environment (RE) and the virtual environment (VE) by providing computer-generated information on the user's view of the real environment. Virtual information can be texts, rendered models of CAD data, or volume data reconstructed from the scans of medical devices.; Vision-based tracking systems are widely used for AR because they do not require additional tracking hardware devices and they can provide accurate registration. However, for these tracking systems, the operating range is restricted to the areas where a minimum number of calibrated features are in view. Partial occlusion of the scene, even when the area of user's interest is in view, may cause failures in tracking.; Tracking over wide ranges can be achieved by dynamically and automatically calibrating unknown features, as they are needed. The calibration is deferred until required as the notion of “lazy evaluation” in Algorithms. With dynamic, automatic, and deferred calibration, a user does not have to predict the possible tracking area, placing and calibrating features ahead of time. Rather, the user starts the system by tracking with a small set of calibrated features. As the user expands to new areas, system calibrates unknown features as they appear in view. Once calibrated, the features can be used as tracking primitives to compute the camera pose. Thus, tracking range can be extended to unprepared environments. The use of natural feature tracking enables tracking range extension even to areas devoid of artificial fiducials, allowing for tracking and virtual object overlay in natural environments. Extending tracking range to a wide area demands reduction of propagated errors because the practical aspect of camera-pose tracking and feature-position calibration involves noise and errors. This thesis describes robust extendible tracking system that removes or reduces noise and errors. |
