RazorEddie wrote:
The ramp controller uses a pic micro *spit* programmed in C. I don't like pics - they are too noise sensitive and often rather buggy. The glue controller will probably be an Atmel AVR or AVR32 programmed in C++.
I'm in the process of designing a POC on a portable electronics device to analyze social signals in realtime. I had planned on eventually using a PIC32, but I just recently happen upon a super-awesome ARM Cortex A9 (the i.MX6 Quad) board. It's so amazing that I'm thinking I might be able to create a partially-functional POC right now.
I had planned on using stereo cams (projective geometry FTW!) to generate PCL data which I would pass through an AI filter (SVM of classified culture-specific social signals), but that would take too long. I figure maybe a Haar cascade on 2D data might be good enough, since it's so much easier to classify.
I don't know ARM programming at all. (I'm really a mathematician with some MSP430 and PIC8 experience.) Do you happen to know much about coding those 3D and 2.5D GPUs? Could I assign each core to a particular cascade to look for results?
I'm using OpenCV for the desktop test (linking it with OpenKinect to simulate the PCL data) but am having trouble getting the Haar classifier to get good results. Do you have experience with other tech that might be helpful?
Thanks!
_________________
Dum vita est, spes est.