Page 1 of 2 12 LastLast
Results 1 to 10 of 16

Thread: Face detection

  1. #1

    Default Face detection

    Can I do face detection with a microcontroller. I haven't decided on any particular model. Since I've heard quite a lot that image processing is too heavy, I was thinking, could I temporarily save a pic in some memory like an sd card and then process it bi by bit.

    The processing need not be real time. So processing bit by bit (if possible) shouldn't cause a problem

    Somebody Please help!!!!!!!!!!

  2. #2

    Default

    The Micro controller should have image processing capability. Look for Blackfin. ( i haven't worked on it though). Ordinary MC are not capable of it. at-least to my knowledge.

  3. #3
    Administrator T-1000 vikas's Avatar
    Join Date
    Feb 2004
    Location
    Delhi
    Posts
    1,922
    Blog Entries
    14

    Default

    I find OMAP3 better than blackfin , its actually competing with ATOM's .

    Go for Beagleboard ,ARM 9 Cortex-8 at 600 Mhz AND DSP running at 400 MHz + would any day help you out.

    Beagle board i highly recommended.Ive used them and they are fun to work with.

    Also they are unbrickable .
    Vikas Patial

  4. #4
    Senior Member T-1000 debu's Avatar
    Join Date
    Jun 2007
    Location
    New Delhi, India
    Posts
    338

    Default

    The OMAP architecture is superior to blackfin. It has a DSP core (some parts have more then one) built in along with the processing core. Blackfin on the contrary only have an optimized DSP instruction set.

    If however (@diffthinkr) is looking to simply perform simple arithmetic (non-floating point) interpolation on an image, he could consider using a really cheap solution. This paper (clickie) could be used as a starting point.

    Hope that helps.

    Regards,

    Debu

  5. #5

    Default

    Which would be the cheapest one that I would be able to use and how much would it cost in India?Where can I get it(preferably in south India?

    @ debu-I was thinking on the lines of OpenCV or something of that sort for the detection. If this is "simple arithmetic (non-floating point) interpolation"
    then please tell me more.


    P.S. Sorry that I'm so specific. I don't think that my dad would let me spend too much on "hobby project"!

    Thank you

  6. #6
    Administrator T-1000 vikas's Avatar
    Join Date
    Feb 2004
    Location
    Delhi
    Posts
    1,922
    Blog Entries
    14

    Default

    They will cost you upwards of Rs 9000 , Blackin is not available as such.

    Beagleboard can do OpenCV face recognition but u need to DSP teh process to get any decent throughputs.
    Vikas Patial

  7. #7

    Default

    Wow!! Would even non real time processing require that powerful systems?
    I was thinking maybe I could store it somewhere and retrieve it piece by piece and process them.I was thinking of this because as I understand it, we are comparing different parts the picture many times with reference pics of different sizes...http://opencv.willowgarage.com/wiki/FaceDetection

    I was thinking of storing and retrieving using something like this.
    http://www.captain.at/electronic-atmega-mmc.php
    I still haven't decided on the processor though.This seems to be costly area so I'm taking my steps carefully!!!
    Hope you guys don't mind my questons!

    Would that work?

  8. #8
    Senior Member T-1000 debu's Avatar
    Join Date
    Jun 2007
    Location
    New Delhi, India
    Posts
    338

    Default

    @diffthinkr: Read the link I provided in the last post carefully. It just uses an AVR for processing the image. If you don't want a GBA camera (which are very cheap and easily available off the shelf here in Colchester) and instead use a MMC with pre-stored images, it would be quite cheap.

    Regards,

    Debu

  9. #9

    Default

    @debu.. 1st off ,I'm very thankful for spending so much time for me.

    Actually, I've come across that paper earlier and didn't give much thought to it due to 2 main reasons.
    1) I couldn't find where I'd be able to get a GBA camera in India
    2) I'm not sure that edge detection would do much help in face detection
    (Not sure, so please correct if I'm wrong)

    Let me explain my idea a little more clearly. Please point out any problems that I may face .... ( I've designed only a couple of designs and this is by far the most complicated)

    I was thinking of using an ordinary webcam. The lowest res output i was able to get in PC was about 10kb.Either i could use the the uc's memory or an extended memory.I've heard of uc's with usb capability( don't know if it is available here,pls let me know).So, I'll interface with that. Extract part of the pic, check it, and the take another and so on. Once the whole process is finished, the next pic will be stored and the process is repeated. This is what I have in mind.
    Don't know the problems involved.

    Please don't feel that I'm ignoring all the advices given to me.But I want to try to build the system with easily available parts and not specialized systems, if possible.

    Please advice and deep gratitude for the help so far!

  10. #10
    Senior Member T-1000 debu's Avatar
    Join Date
    Jun 2007
    Location
    New Delhi, India
    Posts
    338

    Default

    @diffthinkr: Yes, as such you may not be able to use any existing face detection and tracking algorithms on your uC without much work. If you don't want to interface a camera to your uC, and want to provide still images using a MMC (as you stated in the opening of the thread), I propose you perform some kind of demand paging and load the image bits into the uC's RAM in a priority based pattern manner similar to the method followed in this paper (clickie) and in this one (clickie).

    Regards,

    Debu

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •