Having worked a lot with the EEE PC’s before and written custom software I must say…

DAMN THIS IS GOOD, There ain’t anything to work with. If I remember correct this type of EEE pc has 1,6 ghz cpu and 1 gb ram.

If he when could port it to the 900 mhz eee pc 900 it would be even better ! :D

But can it accurately detect power levels over 9000?!

Looks like it’s seeing his nostrils as eyes.

People always underestimate the requirements needed for image processing.
2-D convolution filters require a huge computation process.
OpenCV really makes the development process a lot easier than other development routes which has led to some researchers to stray from the typical Matlab/Scilab platform. I just wish that intel would implement a VHDL synthesis tool into OpenCV to exploit FPGAs concurrency.

Ryan, you make an excellent point, the fact that the Atom CPU, with its near total lack of concurrency, would benefit greatly even from a small and underpowered FPGA running some fast concurrent multiplies and adds to take the convolutional load off the CPU. Although it’s a nice portable platform for a wearable computing system, including connectivity and decent battery life, the power is just not there for truly realtime CNNs at this point.

I’m no expert at any of this, but I’m a long time coder and graphics enthusiast since the commodore days. Would any of this be suitable to offload to a GPU rathern than FPGA? I recently looked into nvidia’s CUDA, and it appears to support OpenCV.

I can see a future when every cop/airport security/rent-a-cop is wearing one of these things and if you see it off they haul your ass off to jail.

hi christopher,

nice nice nice! i was looking for a solution for face recognition, but didn’t find anything open source. will you share you code?

right now, i am using opencv for basic face position recognition (not detecting who is who).

pat

I think you could find something comparable in size with more processing power. Have you seen the nvidia s series of devices designed for small form factor? Here are a couple that could pack some punch…

http://www.newegg.com/Product/Product.aspx?Item=N82E16856110057

http://www.logicsupply.com/products/ipx7a_ion330

Cool beans. By no means a new concept, for example, the SixthSense videos demo another take on it. But there does seem to be a distinct shortage of open-source libraries for this, OpenCV notwithstanding.

The idea of running 6FPS on a VST HMD bugs me though. A VST HMD is annoying enough to wear for long periods already. 6FPS would be “good luck walking to down the hall without hurling or smashing into something” territory. If that’s all the Eee can manage, then it shouldn’t be used as the source. Yeah, it makes the system “portable”…but the result probably isn’t practically usable.

Finally having some useful augmented reality software is gonna be pretty awesome. I’ve built two wearcomps in the past and ended up dismantling both because there was an utter lack of software for them.

It’s a just a shame that OpenCV’s Haar cascade classifiers are so awful. You can see in the video that every time it detects Christopher’s face, it uses a different sized window, sometimes cutting off bits of his face, etc. I’m amazed that the face recognition component works at all with such inconsistent inputs.

not sure if my little biscuit box computer is “wearable” (size, power) but i get good speed under linux with a rt-kernel.

http://www.workinprogress.ca/biscuit-box-computer/

pat

@Christopher

With regards to a portable CUDA solution, have you given any thought to using NVIDIA ION ( http://www.nvidia.com/object/io_72770.html )?

As for using FPGAs, perhaps one of these devices might be of interest if you don’t mind using ARM:
http://www.embeddedarm.com/products/arm-sbc.php
http://www.armadeus.com/english/products-processor_boards-apf27.html
http://www.actel.com/products/smartfusion/default.aspx