Hi, I just found this project and it is really amazing!
I also found a project that implements opencv object recognition in actionscript
( flash ). But like OpenCV it only takes .xml for input.

Did you guys make the conversion?
Or does anyone knows of any other hand recognition .xml I could use?
OpenCV comes with face, body and eyes .xml. But no hands.


Link with news about the flash project in English
http://drawlogic.com/2009/03/16/as3-version-of-opencv-for-facial-detection-web-c\
am-object-detection-head-tracking-3d-library-integration-and-more-with-open-comp\
uter-vision-library/comment-page-1/

Link to original flash project (Japanese)
http://maaash.jp/category/marilena/

Thank you very much, and sorry for bad English.



--- In handvu@yahoogroups.com, "didgy666" <didgy666@...> wrote:
>
> Hello,
> If you remove the features that OpenCV can't handle, is the detection
> still acceptable? I did create a couple of .xml cascade from scratch,
> but with poor results, that's why I was so interested in your
> conversion work. Can you send me your results please. I'm also curious
> about the setting you use to create your own .xml, maybe that's where
> my problems are coming from. Here's what I've been doing.
>
> 1-Collection between 850 and 1250 of positives pictures taken with my
> webcam, size 1600 x 1200
> 2-Collection of 1200 negatives random pictures from my own collection,
> size 3072 x 2304
> 3-All image have been converted to bitmap and they are all sequential
> (img0001, img002...)
> 4-I mark every positive images with objectmarker.exe, witch create
> info.txt that look like:
> rawdata/img0001.bmp 1 409 197 889 958
> rawdata/img0002.bmp 1 483 150 846 996
> 5-I than create the vector.vec file with createsamples.exe, using
> setting like this
> createsamples.exe -info positive/info.txt -vec data/vector.vec -num
> 1100 -w 27 -h 16
> 6-Now the long part, using haartraining.exe, I created the different
> stages of the cascade, with these settings:
> haartraining.exe -data data/cascade -vec data/vector.vec -bg
> negative/infofile.txt -npos 1100 -nneg 1200 -nstages 20 -mem 2000
> -mode ALL -w 27 -h 16 -nonsym -maxfalsealarmrate 0.03
> 7-Finally, I assembled all the stages to create the .xml cascade with
> haarconv.exe, using these setting:
> haarconv.exe data output.xml 27 16
>
> Not once the program has reached stage 20, the best I did was 14 with
> a open hand. Picture of L shape (thumb and index), only reach stage
> 10. I'm I doing something wrong? Is the picture to big? I heaven used
> photoshop to modify the pictures a little bit(reduce noise, sharpen,
> level for color), with similar results. Assuming that the pictures are
> use internally in gray scale, I decided to try to convert it myself
> for the best gray scale conversion possible, didn't help. If anyone
> knows why it produced such poor results, please tell me.
>
> didgy
> didgy666@...
>
>
> --- In handvu@yahoogroups.com, "Mathias Kolsch" <mathias.kolsch@>
> wrote:
> >
> > Okay, I wrote a converter from .cascade to .xml. The catch is that the
> > current .cascades use features that OpenCV can not handle, hence the xml
> > will not work out of the box.
> >
> > Since I got everything up and running on my end again, I can
> probably create
> > you detection cascades in XML style quite quickly. (Takes a couple
> hours
> > upwards, depending on template size and number of images.) What I would
> > need is two sets of images, one with positive examples, and one
> without any
> > occurrences of the object in question. I have in the past used about
> > 100-200 positive examples and 300-2000 "negative" images.
> >
> > If you're interested, let's coordinate.
> >
> >
> >
> > Mathias.
> >
> >
> >
> > _____
> >
> > From: handvu@yahoogroups.com [mailto:handvu@yahoogroups.com] On
> Behalf Of
> > didgy666
> > Sent: Thursday, November 13, 2008 10:41 AM
> > To: handvu@yahoogroups.com
> > Subject: [handvu] Re: Converting .cascade to .xml HaarCascade for openCV
> >
> >
> >
> > Good! I would love to see and use your work. I've been spending a
> > ridiculous amount of time creating xml cascade in the past few months.
> > It's such a mind numbing task... Since I'm working on a school
> > project, I hope it will be out in time... If it's possible, could you
> > send it directly to me? I would really appreciate.
> >
> > Didgy
> >
> > --- In handvu@yahoogroups. <mailto:handvu%40yahoogroups.com> com,
> "Mathias
> > Kolsch" <mathias.kolsch@>
> > wrote:
> > >
> > > The HandVu cascades implement a few more features than the OpenCV xml
> > > cascades. That's why it's not generally possible to convert from
> > .cascade
> > > to .xml. The other way around works theoretically. That said, I'm
> > writing
> > > a converter from .cascade to .xml at the moment myself, but it
> will only
> > > translate cascades there were built with a subset of HandVu's
> features -
> > > those compatible with OpenCV's xml files.
> > >
> > > Once done, I intend to publish an updated HandVu version that also
> > includes
> > > a Matlab adapter so you can run the detection from within Matlab.
> > >
> > >
> > >
> > > Mathias.
> > >
> > >
> > >
> > > _____
> > >
> > > From: handvu@yahoogroups. <mailto:handvu%40yahoogroups.com> com
> > [mailto:handvu@yahoogroups. <mailto:handvu%40yahoogroups.com> com] On
> > Behalf Of
> > > didgy666
> > > Sent: Tuesday, October 28, 2008 12:46 PM
> > > To: handvu@yahoogroups. <mailto:handvu%40yahoogroups.com> com
> > > Subject: [handvu] Converting .cascade to .xml HaarCascade for openCV
> > >
> > >
> > >
> > > Hello,
> > > I wonder if anyone knows if it's possible to use HandVU's .casscade
> > > file with openCV HaarCascade functions. The HaarCascade functions want
> > > to have an .xml file, not a .casscade. Since they both implement
> > > "Rapid Object Detection with A Cascade of Boosted Classifiers Based on
> > > Haar-like Features" approach, there should be a way to do a
> > > conversion. Looking inside both type of file, I notice a lot of
> > > differences in the structure of the data and not just because .cascade
> > > file or not build with three like structure. I tried o translate by
> > > hand, but I realized that .cascade had more parameter than the .xml
> > > file. Can anyone help?
> > >
> > > Thanks for your time,
> > > didgy
> > >
> >
>