Jul 11, 2011

Pandaboard face detection performance test

We did a simple performance test of OpenCV's haar finder face detection with Pandaboard on Ubuntu and openFramewroks.



 The Ubuntu is: 2.6.35-903-omap4 #21-Ubuntu

Compiled with Code::Blocks and openFrameworks with the ofxCvHaarFinder addon.

The source of the test can be downloaded here.

Here is a video of the screen with webcam capture resolution 320x240.



And here with webcam capture resolution 160x120.



[EDIT]

Here is a test capturing at 160x120, but with the haar finder set up for a minimum blob size of 16x12px (the bigger minimum blob size, the faster the detection, with the loss of detection for small/distant faces).
The code is rendering the result via SDL on the framebuffer. The gdm was shut down before running the application.

5 comments:

  1. Hi, Great post!
    I am building a similar system, could you please advice me what's a compatible camera for ubuntu/pandaboard/openvb, Thank you

    ReplyDelete
    Replies
    1. Sorry for the late reply...

      At the moment I'm working with two great cameras, both of them should have no problem with your panda:

      - a Sony PS3 eye webcam (see patches available to control the camera parameters)
      - a Logitech C910 webcam (wich properly tuned may give you HD video feed)

      Delete
  2. Did you install some specific drivers for those cams ? Or is it working out of the box ? Many thanks

    ReplyDelete
    Replies
    1. Hi Quentin,

      That depends on your kernel. As far as I remember for us it worked out of the box (try installing guvcview from the repos and test the camera). However, if you want more functionality with the ps3 eye on linux, have a look at this: http://therandomlab.blogspot.com/2011/12/ps3-eye-driver-framerate-patch.html

      Delete
    2. Hi David,

      Many thanks for your reply. I tryed with the 3.k kernel (on kubuntu) and it didn't work. I'll try with an old ubuntu distro (Maverick for example ;)). I let you know.

      Delete