As i was daydreaming and reading the newspaper i stumbled on an article on augmented reality applications.
I've seen some example movies on youtube as well on this augmented reality.

In this one the augm. rlty. browser Layar detects celebrity houses and this guy finds Brad Pitt somewhere in Amsterdam. (watch to the end to see brad pitt)
http://www.youtube.com/watch?v=6BybSbY9NR4

Now i do not see the fun in augmented reality myself, since it is not much more than google maps or whatever. (as it works with the gps)

What i'm about to say might be far stretched and i don't even see a practical use for it.
I was just wondering if anyone here has the knowledge to provide the insight if it would be theoretically possible to create an app to actually analyze objects which are snapped by the camera of the N900 or filmed by it. For example, when you film/fotograph a football, that the application would be able to say: object: football. brand: nike. available at: www.nike.com.
Since maemo is open source this would be THE operating system to think of.

this way you could create a way to analyze someones face or what might be even cooler, to capture a text on paper, and then recognize the characters to transfer this to a word document.
just as some scanners have text recognition.

Don't quote me on this but object recognition on the level you are talking about is afaik still on the research table. Very complicated to identify things like that based only on images, especially intelligently without building a database of "known" objects to compare with. Barcodes and rfids are two ways of identifying stuff (a camera can be used to identify barcodes, think there is something like that for s60 for instance) but that's beside the point...

Text recognition on the other hand sounds semi-feasible to me, but don't know what technologies there are. Most documents are really available electronically today anyway...?

thing is, the stuff the professor writes on the schoolboards isnt always

Good point. There are probably several other cases where it would be handy.

Actually I had another thought about the text-recognition in conjunction with AR. Suppose you are abroad and don't know the language. There are a number of occasions when you would need to read signs, menus and what not. How about if you could point your camera at the text and get a translated version of it in your display?

Now, text translation is not easy either but there are solutions (google, babelfish etc) and they will only get better. What is remaining is basically to process the image, identify what is text in it and process that into "real" text as input to the translation algorithm. There are techniques to identify 'glyphs' in an image and associate them with virtual objects. I wonder if that technique could be refined to treat each letter in an image as a 'glyph' and associate it with corresponding "real letter" to build up a text in a software..

Does anyone have thoughts on this..?

Semi-feasible was probably a fair assessment:

http://en.wikipedia.org/wiki/Optical...er_recognition

I actually think a text-translating AR app is not entirely far-fetched for "easy" signs with say roman letters and simple fonts (personally I think this would be really cool and actually usable..). Once you can capture text successfully you could not only translate it or transfer it to a document but also do things like automatically search for it on the web and whatever..

GOCR, a GNU open-source character recognition program in development:

http://jocr.sourceforge.net/

I think there could be a future for this.

http://en.wikipedia.org/wiki/Automat...ion#Algorithms

That one caught my eye as well.

http://www.bitquill.net/blog/?p=119

intel showed off a demo of street sign translation close to the beijing olympics, but that was powered by a backstage stack of computers doing the actual translation (not that google could not provide such a backend with its translation service, just get from image to text, and dump it to them over a data connection of some sort).