Fast Image Recognition using FPGAs

Posted 13 Feb 2005 at 02:04 UTC by steve

A new Embedded.com article discusses an interesting technique in which high speed parallel processes for image recognition were moved into FPGA hardware. A small board, called the Intelligent Camera, was built as a test project. The camera recognizes traffic signs. One step in the image processing required sorting 9 data values for each pixel, resulting in over one million operations per frame. The sorting operation was done using an algorithm that allowed the FPGA to completed the sort for the entire image in parallel in just 2 clock cycles. To quote the designers, "for certain tasks even the world's fastest DSP appears slow in comparison to this solution".

Where's waldo, posted 13 Feb 2005 at 14:30 UTC by The Swirling Brain » (Master)

Let me be the first to say, "Wow, very cool!"

Now, what kind of games could you play with this thing: where's waldo, or better yet, hold it up to a mirror to see if it can recognize itself, etc. :-)

It looks to me (no pun intended) like it works the same way as voice recognition where you give it say 5 words (pictures in this case) for it to recognize and it give you back a yes or no type qualitiative response. So you could give it pictures like a fuzzy cat and if it "hits" it could give an "awwww, here kitty kitty" response. Or if it hits on a picture of a wall plug it gets ready to charge its batteries, or if it hits on a picture of a terrorist, it goes into stereotypical killer robot mode and blows everyone away! (ok, yes I'm a little weird)

The system seems cool, but it doesn't really seem to be a good solution for navigation, though. So I don't expect to see one of these on top of a grand challenge hummer. I would think this would be something like you'd use on an assembly line for seeing the difference between a good widget and a bad one and sorting them to their correct tub.

See more of the latest robot news!