Augmented Reality (AR) is the practice of adding computer-generated information to something you are experiencing already. Often, this involves injecting computer-generated images into a live video feed, and there's some very impressive stuff out there which sees computers spotting special tokens (fiduciary markers is the jargon name) and overlaying images or videos.

However, it turns out that there is another, simpler way of letting the computer know what you are looking at: if you can specify the exact location of the camera in relation to what it is viewing, then you can go ahead and enhance things to your heart's delight, without having to go around tagging everything with markers.

As an example of this, I've found that augmented reality has become a lot more, err, real, thanks to my new phone, an HTC Dream. It contains some handy embedded hardware that means it knows exactly where it is and where it's pointing (the GPS tells it where it is on Earth, the accelerometer tells it which angle it is pointing relative to the surface, and the compass tells it which direction it is facing relative to the poles).

One application that takes advantage of this is Wikitude AR - it will search publicly available sources of information, and then overlay their position on top of the camera feed from the phone, as you can see from these screenshots taken while I was looking out of our office windows:

Then, there's the truly amazing Google Sky Map: you simply point your phone at the area of sky that you are interested in, and it will overlay constellation information. You can also type in a search term (e.g. Venus), and it will give you a handy arrow and target, helping you to swing your phone round until you are pointing right at it. This is extremely useful for any other parent who has been pestered by their child wanting to know what that bright star is called...

Location-based AR has also been implemented for museums and galleries. For example, this movable screen in the Allard Pierson Museum.

But, for me, the most amazing example of this technique that I have come across is from a small research institute in Italy: scientists at the Technologies of Vision department of the Fondazione Bruno Kessler have created a system called Marmota: a way of injecting extra information into photographs by automated terrain analysis.

I got in touch with Michele Zanin from the FBK, and sent him a photo I had taken which had been automatically geotagged by my phone. He explained that the rolling hills of the the South Downs are much harder to match than the more prominent crags of the Italian Alps, but nevertheless managed to tag the peaks visible from near Foredown Tower:

Not only that, but once the terrain had been identified, he was also able to send me a file which overlaid the photo on Google Earth. Click on the screenshot below to download the kmz file yourself.

And peak analysis is only the beginning: the group has already managed to insert road information from OpenStreetMap, and are talking of using the system to automatically track snowline data by aggregating photos, as well as 'crowd sourcing' abstract ideas such as 'attractiveness' by bulk analyzing Flickr comments for geotagged images. You can see more of Marmota by looking at their photo explorer or their blog, as well as reading their presentation (PDF, 6Mb).

All in all, location-based AR looks like a very exciting field to watch.