The Status Quo Of Computer Vision

prostoalex writes "The Industrial Physicist sums up the recent advances and developments in the world of computer vision. They mention an application for human-computer interfacing using a Webcam, Philips Research Lab Seeing with Sound product, which augments vision for visually impaired, as well as various frontal face detection applications."

Re:Don't forget the DARPA Contest

[Animats]

That's not doing much for computer vision. Most of the action in computer vision right now involves "homeland security" applications, real or imagined. The killer app for computer vision seems to be Big Brother.

Re:Artificial intelligence in under 20 years

[tpearson]

We already have basic AI - we have UAVs that can plan their own route and complete their mission completely autonomously. Most commercial robots that are being released have enough AI to determine where they are, what their goal is, and how to perform that goal. In my opinion, that is definitely "basic" AI.

Some of the posts on here are getting a bit vapor

[sielwolf]

Maybe it's some sort of technophilia but some of the posts on here are just pure vapor. Sure, there have been some great advances in computer vision and pattern recognition... but have some of these posters on here ever done any research in the area? Hell, most face recognition goes back to Fischer's 1936 iris data set and primary component analysis... not quite Wintermute stuff.

Too often vision projects find speedups by sacrificing one or another components. For instance, you can get some great face recognition with PCA... as long as the person's face is immobile. Tilt your head slightly or rotate too much and the system has no clue.

I'll admit, there is some killer work out there. But not of the full-blown "20 years and we will all have robotic man servants" thing. Keep the hype to a minimum.

Re:Some of the posts on here are getting a bit vap

[t]

No kidding. I'm personally quite disappointed with state of the art speech to text, computer vision, etc... Much of it has gone largely unchanged for years, optimizations here and there is about it.

I think at some point we went down a path which will never lead to the solutions we expected to have by this time. And the reason we can't get off the current path is because of the way the tech culture is, you always have to publish an extension to previous work with copious references.

And its not even the big stuff, look at spell checkers and grammar checkers. Are there any that can tell correctly spelled but misused words? Affect/effect? There/their? How about something easy like made and maid?

high speed high res cameras

[hyperventilate]

I was stunned by how OCR went from "impossible" to "Trvial" and all that changed was moores law making high res scans available in memory in a typical PC. Expect many vision problems to fall by the wayside with new 240 Frames per second 3 megapixel cameras [fast-vision.com]. (Don't save THOSE movies uncompressed!) See the Sensor Spec [fast-vision.com].

Re:Artificial intelligence in under 20 years

[deblau]

Sorry, I have to disagree. I think you have some very good ideas about what could possibly be automated, but the devil is in the details. One of the biggest pitfalls is assuming that a computer would somehow be 'smarter' than a human, just because it can perform fast calculations. For instance, you claim that an AI could turn a book into a movie. Great idea, but I know I can't do that myself, and I like to think I'm pretty sharp. I'm also pretty sure that most of the people developing AI can't either, or they would have been screenwriters for a living. Who exactly will train this AI to write movies, and what kind of skills will they require to do so? These questions have pretty vague answers, and computers don't tolerate ambiguity very well.

I also think that your claim of progress toward hard AI because of vision advancements is a little misleading. A true AI will gather perceptual input from a variety of sources in order to get the most accurate representation of concepts. For instance, a rose isn't really a rose if you've never smelled one; instead, it has the same emotional impact as a strawberry, a pile of vomit, and a face -- it's just another hard, lifeless image. Of course, if you can build me a machine that generates a good, emotional screenplay from some words on a page, I'll buy it from you for $100,000,000 and consider it a good deal. Heck, I'll turn out blockbuster movie remakes of classic literature by the thousands and make that figure 100 times over.

Game advancements don't really help out AI either. What they do help is expert systems, which is a related field, and one which many people confuse for hard AI. The basic difference is that the input model for an expert system is generally limited to a single topic, whereas a general AI trains on any input it can perceive. There are many "AI" projects out there which train to recognize faces, and similar tasks. These projects are really expert systems, since they'll never be good for anything beyond face recognition, or for whatever limited task they train. You wouldn't ask a face rec program why good-looking people succeed in politics, now would you? But you would ask your Marketing major buddy.

To wrap up, I think expert systems is a thriving field, and that for many problems an expert system will be good enough. I wouldn't hold my breath waiting for a real AI, though.