[PD] Computer Vision with Pure Data
jonastikov at gmail.com
Fri Dec 5 14:24:26 CET 2014
Thanks for your reply!
I have also looked a bit at the Leap Motion, but what I would like to know
more about, is something that could be implemented on a "smart device"
using the in build camera as sensor. For the same reason the reactable
solution is a bit outside my scope.
I have already looked at the openv_contours_boundingrect, but it is marked
"deprecated" in the help file, so I thought I would not spent time on
making that work. Any thoughts on what the problem is and if it would be
possible to use anyway?
Your suggestion on using pdmd is intriguing, but I cannot understand how to
make use of it (perhaps since I have hardly used pdmd at all).
With the different setups I have it seems possible to make solutions to
detect x-axis and depth. With the y-axis I have a problem since I am also
detecting the arm, which disrupts the motion tracking. At the moment this
is my biggest issue.
Any suggestions except waring long black sleeves are most welcome!
Hints about other pd projects detecting hands with web cam would also be
On Fri, Dec 5, 2014 at 12:39 PM, peter venus <news at petervenus.de> wrote:
> Hello Jonathan!
> as i understand it, you want to track only a hand.
> i don´t know if you know about the Leap Motion Controller.
> This one is specialized to detect hands and from what i read in your mail,
> could really fit.
> i have to admit, i havent played with it myself yet...but there is support
> for pd available .
> if you want to stick to the pix_openCV library, have a look in the
> if you leave out the physical modelling stuff, it is quite usable to track
> people, maybe hands as well.
> If you want to or have to stick to a camera based tracking system, i could
> see some possible improvements:
> as you say
> > The idea is to detect hands from an up-facing webcam in order to
> simulate a
> > camera on a touch pad.
> if you really want to simulate the hand being tracked on a touchlike
> surface, why not add a semi-transparent plexiglass layer and place a camera
> underneath it, similar to the concept behind the reactable.
>  https://www.leapmotion.com
>  http://puredatajapan.info/?page_id=1514
> Am 05.12.14 01:12, schrieb Jonathan Aardestrup:
>> I am working on a university project using Pure Data for computer vision
>> order to investigate possible advantages or disadvantages in a music
>> related interaction context.
>> The idea is to detect hands from an up-facing webcam in order to simulate
>> camera on a touch pad.
>> I've spend some weeks to get familiar with the objects in Gem namely pix
>> and pix_opencv, but even with objects such as pix_opencv_bgstats or
>> pix_opencv_bgsubstract, alone or in combination with other objects, have I
>> been able to get a clear binary image to detect motion from.
>> If you could possibly have a look at my code and give me some hints to how
>> I could improve it I would *really* appreciate it! Other suggestions are
>> also most welcome!
>> What I want it to do is detect hand (preferably without arm?) and give me
>> the following readings:
>> x-pos, y-pos, x-direction, y-direction, x-acceleration, y-acceleration
>> blob size. x and y being for the center of the detected hand. Multiple
>> blobs would also be great but I can't really understand how to use it so
>> I've attached one out of many attempts for the right solution.
>> I am using some objects from pix_opencv library which (in case you don't
>> already have it) can be found here:
>> The best
>> Pd-list at lists.iem.at mailing list
>> UNSUBSCRIBE and account-management -> http://lists.puredata.info/
> Pd-list at lists.iem.at mailing list
> UNSUBSCRIBE and account-management -> http://lists.puredata.info/
-------------- next part --------------
An HTML attachment was scrubbed...
More information about the Pd-list