DSP + OPenCV

Hi all,
I build a OE image for the Beagle Board, OpenCv included as well, using [1]. I tried a OpenCV example on BB (Facedetect) and i think it was ok! I can’t use for the moment any TV so i’m using minicom from my laptop to send and execute…So i runned this example(i applied it on one photo) on my laptop and it detected face in less than 1 second, instead on BB it detected face in 8 sec more or less…So i would like to use the BeagleBoard DSP to reduce this detection time.

  1. How can i use the BB DSP? Does exist a user-guide to do it step by step?
  2. Supposing i could activate the DSP, how can i connect it to my example-program? Do i have to use a particular compiler for DSP?If so where can i get it included a tutorial or user guide that explain me how to use it?

Thanks in advance

I profiled facedetect and it spent ~95% of it's time going through the 1MB xml file and applying the cascades. You could try to trim down the xml file to get a faster, but less accurate match.
If you do decide to use the dsp, have a look at dsplink.

regards,

Koen

Thanks K

2009/1/18 Koen Kooi <koen@beagleboard.org>

Like this one?
http://elinux.org/BeagleBoard/DSP_Howto

Hi Orazio,
can you give more details on how you included the OpenCV in the image
so that other member (like me) can do the same.
Regards.