Browse Theses

In Augmented Reality (AR), a user sees a virtual world as well as the real world. The virtual world should be correctly aligned with the real world by keeping the user's pose in both worlds exactly the same. An AR system should have the ability to track the user's movement accurately in a full work-area of an application. Fiducial tracking is an attractive approach towards the registration problem in video see-through AR due to the availability of digitized real scenes to an AR system. However, most of the current fiducial tracking AR systems have very limited tracking ranges (in most cases, small desktop workspaces). None of them seems to have scalability to support a wide range of applications from small-scale desktop tracking to large-scale full room tracking, and a wide range of views from detailed views at arm-length distances to whole views at far distances in an application.

For scalable fiducial tracking, the following problems should be solved in addition to the currently developed fiducial tracking AR systems: (1) Fiducial design for a wide detection range. (2) Robust and fast fiducial detection. (3) Fiducial cluster identification.

To solve these problems, the following were developed in this dissertation: (1) A multi-ring color fiducial system. (2) A multi-phase multi-threshold fiducial detection method with a 1-D segment filter. (3) A two-step fiducial cluster identification with an angular projection-invariant (API) graph and a view-based orthogonal affine transformation (VBOAT).

In this dissertation, a simple and low-cost way to achieve scalable AR is developed. Since multi-sized and duplicated fiducials are allowed, there are many available fiducials that can be used to build a large work-area in one application. It is also possible to extend a work-area with multiple local fiducial groups in a large place or even different rooms or buildings. This approach virtually eliminates the geographical restriction of a work-area.