“Eye-Phone” Image Recognition System Coming?

by Eric March on May 1, 2008 at 11:26 am

I’ve seen some novel ways to use GPS technology and the Internet lately — especially down yon Android-way with the Enkin demo. We have also seen and reported on ViPR, a fascinating new way to use your iPhone’s camera for image recognition in order to look up books, movies, CDs, and so on, wirelessly over the internet from wherever you are. So what would happen if you combined all of these technologies, added the ability to learn from its environment, and feed it all back to you in an augmented reality display through your mobile phone and its camera?

You may very well get Superwise Technologies‘ “eye-Phone.” Designed by Ernst Pechtl and Hans Geiger, this regional winner of the ESA Technology Transfer Program’s Satellite Navigation Competition combines your phone’s digital camera, satellite navigation localization services, advanced object recognition, and an Internet connection to retrieve relevant information about what it’s seeing to present an augmented reality display on your phone’s screen, giving you the details of what you’re looking at as you’re looking at it, with an on-screen option to select whatever it’s seeing to obtain even more information about it.

Geiger and Pechtl state that eye-Phone uses an artificial intelligence core known as Apollo, which is capable of analyzing and recognizing objects from within images, and can “self-learn,” requiring only “a short and very simple training session” before it is capable of identifying “any object in the world.” “It could be a building, a mountain, a tree, plant or a special event such as a local festival,” Ernst said.

While I have no doubt that the technology is certainly possible, as we have been shown by the practical demonstrations featured in the links of the first paragraph, I am having trouble with this “short and very simple training session.” The idea that it can “identify any object in the world” strikes me as completely specious — there are a lot of objects out there. An entire planet full of them, in fact, from dust mites to mountain ranges. I am thinking here that the wording is highly ambiguous, and that “any object” really means “any significant, documented image.” Certain things I can fully understand; it could conceivably identify geographical landmarks, plants, birds, animals, etc., and consult reference material on geography, botany, ornithology, zoology, and so on to give you the relevant information you are interested in. Where it trips me up is this whole festival thing. While I understand that there is an AI at work here, I can’t quite work out how it could identify, say, a Caribana parade, given that such festivals are typically marked by no specific traits other than wild, colourful (but always different) costumes and upbeat Soca music.

Nevertheless, Superwise are planning on pitching this technology to mobile OEMs in an effort to position their product as a premium subscription-based service that would work as a combination of embedded hardware and network-based software located on a central lookup server. I do find the technology quite fascinating, even if this one doesn’t track in realtime like the Android demo, and it does have the potential to be an invaluable resource, especially for tourists wandering aimlessly around unfamiliar territory. I do question whether this will catch sufficient interest from OEMs, however, especially because I can’t help but think an all-software solution to this technology will be just around the corner. Enkin isn’t too far from it, after all, so it isn’t hard to imagine eye-Phone-like functionality being added to it, and ViPR functions a lot like eye-Phone in its methodology. Furthermore, with numerous GPS solutions coming to the iPhone, there is bound to be a flurry of development that can make use of them for things other than simple map navigation.

Either way you slice it, all of this out-of-the-box thinking and development is really going to heat things up for this next generation of mobile development.