Release of HAAR classifier trained for hand gesture recognition

hand detection

hand detection

The release of HAAR training classifier for hand gesture detection

This is the last post of the GSoC hand gesture detection series, and the trained HAAT classifier is finally released here.
This is only a short announcement, but the classifier in .xml file can be reached here in
The release contains two .xml classifier files, respectively they are for CLOSED PALM and FIST detection.

Hope this is useful.

Author: Andol Li

A HCI researcher, a digital media lecturer, an information product designer, and a python/php/java coder.

14 Comments On “ Release of HAAR classifier trained for hand gesture recognition”

  1. Hey Andol, great blog!! I’m wondering what the performance was like for this hand gesture recognition? Did it work, more or less? I’m working on a project for hand gesture recognition and right now im taking this haar classifier approach. I’m going to look at your code more, but I’d love to get your opinion on how well it worked. Thanks for the blog and interesting articles btw! 🙂

        • approximately 800 positive samples and 1100 negative samples, have tried more samples up to 2000-3000 but did not get noticeable improvement

          these samples were hand selected and cropped, each containing good quality target object features.


          • what command did you used to create these ? i having a difficulty in training. there are always too many false detection.

            these are the steps that i do:

            cropped hand images 100 x 100 greyscaled (858 pos , 597 neg)

            used this command to create samples
            createsamples.exe -info positive/info.txt -bg negative/infofile.txt -vec data/vector.vec -num 858 -w 20 -h 20

            then trained using this
            opencv_traincascade -data data/cascade -vec data/vector.vec -bg negative/infofile.txt -numPos 858 -numNeg 597 -numStages 11 -w 20 -h 20 -featureType LBP

            what do you think i done wrong ? wish you could notice this as soon as possible

  2. Hey Andol! One more quick question, do think it would be beneficial to create two separate classifiers– one for each hand (i.e. open palm right hand and open palm left hand)? I suppose hands are pretty symmetrical, but I wonder how haar training handles such discrepancy.

    • @Robert
      actually a separated xml file is not very necessary, coz as you said the hands are mostly symmmetrical, that means it will be easier to do the detection with a flipped image.

  3. Hello sir, I am working on a hand gesture recognition project confused about whether to use a haar classifier or any other image transformation technique for the detection. Any suggestions would be valuable. Thank you.

  4. Hi, fiirst of all, you’re doing a great job. OpenCV lib is a amazing library but it’s not so easy to use.

    I’m working in a project like this hand gesture but with a different approach. Some questions : can I use your haarcascade file? What is the license that you use? Your cascades can recognize the hand and the fingers?

    Thanks in advance. Soon I intend to publish my project that have people with disabilities as target.

Leave a comment
Due to technical adjustments, the comment function is shortly closed and will be re-openning soon. Thanks.

Copyrights 2006-2017 © All rights reserved
Theme Tree2, re-designed by Andol Li, powered by WordPress and Bootsrap
WWW.ANDOL.ME | 浙ICP备15040508号-1
公安备案图标 浙公网安备33010602004018号
Back to top