Overview Of System’s Architecture
Major requirement that our design is expected to fulfill is the support of multiple concurrent users without any serious sacrifices on the response time of the system. The architecture is a client/server architecture, where the clients are the wearable computers of the Mobile Units (MUs) equipped with a wireless network connectivity card. A wireless network with a sufficient number of Access Points (AP) provides connectivity to the server who is responsible for updating the contents of the MU’s database whenever the user is moving to an area about which there is no content available. Graphically, we depict system architecture in figure 1 (see Fig. Archeoguide_System Architecture).
The hardware components of the system include a Site Information Server, the Mobile Units, the Wireless Local Area Network (WLAN) and a Differential Correction GPS reference station, while the software components address the storage, tracking, interaction, distribution and rendering needs of the overall system. The Mobile Unit software has been written in C++ (for speed) whereas the server components were written in Java (for maximum portability.)
- General Description
In order to integrate the virtual objects into the real environment, i.e. augment the user’s view; we need to determine user’s exact position and direction of view. There is a large number of tracking technologies today that offer position and orientation tracking with high precision and low latency . However, none of the available systems is suitable for outdoors usage with sufficient precision as required by ARCHEOGUIDE. Integrating different tracking technologies into one hybrid system seems to be the most promising approach. A first rough positioning of the user is given by using the GPS and compass system. For exact tracking we use imagebased techniques. The only additional hardware that we need in order to perform the vision-based tracking is a tiny off-theshelf camera attached to the user’s HMD. The system can determine the user’s viewpoint based solely on the video image. This brings us to the concept of “image registration”, something we used in ARCHEOGUIDE and which will be discussed next.
- Describing Image Registration
This method for determining the user’s viewpoint assumes that a sufficient number of calibrated reference images are stored in the database, indexed according to the spatial coordinates from which they were taken (successive images overlap to a good degree). The areas from which these reference images are taken comprise the so-called “selected areas” or simply “augmentation view-points”. These areas must be carefully selected in each installation site of the system so as to give a good overall sense of the site (should have open view of most of the site’s important areas and cover them well). The method then performs a comparison between the images that the user sees via the camera attached to the HMD, and a number of reference images in the database whose indexes are close to the coordinates provided by the GPS and compass devices. The matching is performed by considering the image as a whole (global method) instead of identifying landmarks in each image (local method). When the two images being compared overlap in approximately 30-40% or more of the area, the method can correctly compute the warping transformation of one image to the other. This (invertible) transformation is then used to provide accurate head-pose estimation since the coordinates of the database image are known. The entire process is shown in figure 2 (see Fig. Archeoguide_Image Registration). In fact, in the case when there are pre-rendered images of the 3D reconstructions of the monuments rendered from the same point of view of the stored site images, the same warping transformation can be used to transform (fast and accurately) the pre-rendered image and then display it on the user’s HMD thus achieving 3D visualization as well.
(For a more detailed description and analysis of different approaches to Image Registration see Paper_Archeoguide_Sticker et al.)
The interaction of the visitor with the system requires advanced multi-modal interaction techniques since we intend to minimize use of common equipment such as mouse or keyboard. For the first prototype of system however we are developing 3D GUIs based on game-pads or pen-table interactions, on touch screens that change their content as the user moves from area to area. In that way, we treat the user as the mouse cursor on an active image map that is the catopsis of the site itself!
(Didier Sticker et al.)