Tuesday, October 14, 2008

A real “stretch” goal

My ideal machine vision system would function like a human end-of-line inspector, (only more reliably.) You’d show it a few good parts and off it would go, sorting out everything that doesn’t conform.

Sounds like a neural network, only those haven’t yet delivered on their early promise.

Well a team of Brits have decided to go one better. They are going to “
Reverse Engineer the Vertebrate Brain,” (REVERB) and so create machine vision to rival human vision.

Their plan is to use the
SCAMP chip (opens as a pdf) as the heart of an intelligent vision system that would have applications in vehicle navigation, security and automation. For an overview of the project, read this report in ‘Automotive Design and Production

And now my take on it:

They haven’t a chance.

The only thing that machine vision and human vision have in common is that they both start with the capture of photons on a receiving surface. Human visual perception is hugely complicated and not completely understood. In contrast, machine vision, no matter how sophisticated the processor, is just software. It’s the electronic version of gears turning in a machine. It’s just code expressing sets of instructions. There is no intelligence, and without intelligence there is no ability to understand and act upon visual information.

But I’m just a plodding engineer, so what would I know?

No comments: