Since I am presenting this paper, I thought I would give a brief introduction. This particular paper had a big impact within the Augmented Reality community as it was one of the first to do Augmented Reality well. Although the technology is a bit dated (1996), many of the ideas from this paper are still relevant today. As a part of my presentation, I will also provide an up to date view of Augmented Reality. There have been some interesting developments since 1996.

This paper is a little different than the papers we have done so for, as it is similar to a systems paper. However, it serves as a motivating paper to the vision world because the computer vision portion of the research is vital part of the system. This is certainly one potential application that would put computer vision in everyone's homes. Here's looking forward to the future! ;)

Edited by author 10-06-2003 10:26 PM
Some cool stuff, too bad it focused so much on technical issues. It seems that the computer vision techniques they used were extremely primitive, although the real time constraints of 1996 probably account for much of that. Using the cute artificial landmarks whose real loctation was precisely determined beforehand was annoying to a computer vision person -- but perhaps it's a realistic trick for AR? Also, an interesting thing about the application was that quantifying performance was difficult -- I like the trick of using simulation to test an AR system.

I look forward to learning more about current AR systems!

My presentation is also about AR, a bit more recent and purely based on computer vision. It's nice to know how AR evolves through years.

BTW, those who love playing Quake may want to visit http://wearables.unisa.edu.au/arquake/ . I found this is an interesting example of AR.

A few more things... First, videos related to this paper are here:

http://www.cs.unc.edu/~us/web/quicktime.htm

Also, regarding AR Quake. Unfortunately, it doesn't work very well... Although it looks nice on their webpage, a lot of results have been "hand-waved". I think they did it to gather a lot of attention into their research (such as Slashdot).

This paper certainly is very interesting to read.

How does a magnetic tracker work? Is it still in use today?

Landmarks are very important to registration, but looks like they are all pre-defined. How practical would this be in a real-world situation? Would it be possible to use some other object in the scene as landmarks instead of the two-color concntric circular dots?