Wednesday, April 29, 2009

Vault-OS : ThinkBoy recognizes me on camera!

Using example code as a guide, I got motion tracking for video from any input source running last year in C#. I have been using this as my basis for platform orientation and IF laser pointer spotting in the experimental setup in my study.

Last night I was able to stream the genetic memory of Thinkboy to a 4GB compact flash card set up as the second IDE controller between runs of the system and Thinkboy is able to distinguish my face from my wife and children easily. If you read the documentation for the AForge AI open source library you will find ample support for serialization to any storage medium. I was having to train Thinkboy each and every time I booted it up to recognize my face, so it wasn't much use to me until I got it's brain onto an archive device. I have nicknamed the general AI system for Thinkboy (which is really nothing more than the applied code I leeched from AForge AI) as "Integrity" to describe the serialized brain on disk. So if you want to support "Integrity" in your version of Vault-OS you'd do well to set up a separate, atomic instance of a storage device to maintain it so you don't have to share this kind of data with your base cut of the code on your boot flash.

The real frontier for me is gesture recognition and I have been reading the docs of AForge to try to figure it out. I'm starting to get it and I have confidence now that training in gesture recognition won't just be an interesting exercise as a novelty now that I have figured out how to serialize Thinkboy's AForge brain to storage between runs. I have been running the image recognition routines in a separate worker thread and this seems to operate perfectly with the MS SAPI (Speech API library) without a single hinch in performance. My bottleneck in the past was running the COM port control of I2C together with SAPI and getting hideous degradations in performance with more than 1 i2C device online being polled. I seem to have this problem licked now, too, although I have yet to package it all in my little orange box and run it concurrently - I2C network, Speech Control, Visual recognition. That's a big milestone in and of itself. Previous tests were on my P3 board at 233 mhz .NET 1.1, Thinkboy's current board is now a VIA-GPC running at 1 Ghz, which should produce some nice results. I am offshoring the actual monitoring of the CCD cameras to another device with an IP server so this should allow ThinkBoy to concentrate on what it is supposed to do best - THINK and direct everything else in the Vault-OS system.

My baby is going to be uber-win when it is ready for primetime. It's going to be the OStest with the Mostest.


Anonymous said...

treasonholdings said...

autistic kid takes TSA to school:

Anonymous said...

Black Hebrew with interesting things to say: