Note: this video capture was rough, i'll sort a better one out next time. Video first is in debug mode showing the tracking points etc...then in installation mode for rest of vid.

This is what i have been working on part time for the last couple of months. Another particle system; i think that's all i ever build. Thanks to all who made examples and all the guys at openFrameworks.

This is 2 million particles (4 or 8 when mirrored) running on OpenCL/C++ using openFrameworks. It also uses Kinect for input and tracks two separate hand blobs. It runs around 59fps without Kinect and 29FPS with (although that's probably down to me rushing out and buying a Kinect yesterday and coding that part over night to get it up and running nicely, lot's of optimization on that part still to be looked at i guess).

OpenCL on this occasion was faster targeting the CPU compute units (2xquad Xeon 2.xx, ATI 5770, OSX 10.6.x) as the memory transfer to the graphics card is more costly than the compute, however a further step would be to initially transfer the particles to the graphics card then build the VBO there on the card so no memory transfers. Currently i think the particles are 58MB so likely too much for my card to store.

It was fun, and this being my first openFrameworks project it was an easy transition from the very nice Processing framework.

Next step awaits delivery of a 8'x4' chunk of frosted perspex to back project onto, and then we are off to a gallery installation...when i find one!

Obvious problems include further tracking work (i couldn't link a blob tracking library so i just guessed at what would work for tracking and it's decent but gets mixed up often)

Oh and this does track music...it's playing Bjork Unravel but i didn't want any posting issues so i didn't include the audio.

kindest

nik
PolyMath Studios Vancouver