ASL Finger spelling recognition using C# OpenCV wrapper

ASL Finger spelling recognition. 19 identified signs, (currently still working on 7 other signs).

Translasi bahasa isyarat mengklasifikasi 19 isyarat huruf tangan 7 isyarat lainnya masih belum bisa.

- Haar object detection (detect open and closed hand) to determine the position of the box.

- Movement detection (absolute difference). If shaking then reset ROI box, if no movement then detect the shape.

- To extract the hand shape I use skin detection based on HSV or RGB (it depends on the lighting) then convert it to binary image.

- To classify the image I use KNN (look up the machine learning section of OpenCV) I use 100 training images per signs.

paper : http://www.eepis-its.edu/uploadta/downloadmk.php?id=945 (sorry it's in Indonesian)

source code:
http://www.4shared.com/file/2AwWOJ_V/OpenCVLib5.html

Huge credit goes to Heiko Kiessling for creating easy to use C# OpenCV wrapper and helping me (alot) in creating this program.
You can find his code project article here :
http://www.codeproject.com/Articles/28465/Easy-to-use-Wrapper-DLL-for-Intel-s....


Noob guide:

1. Download opencv from here : http://sourceforge.net/projects/opencvlibrary/ (I used the old version of opencv 1.1pre something)
2. Download OpenCV C# wrapper from here : http://www.iib-chemnitz.de/cvwrapper/download.php
(I used the one for opencv 1.1 pre)
3. Make sure that you can successfully run the big sample from the wrapper before attempting to compile/run my program.
4. You may have to edit and play around with the code because some settings such as resolution is hardcoded. May the code be with you.


some rights reserved :)

Category:

Science & Technology

Tags:

• OpenCV
• ASL
• C#
• Computer
• Vision
• Image
• Processing
• KNN
• haar
• cascade
• Hand
• detection
• motion
• image
• classification
• feature
• extraction
• skin
• color
• sign
• language
• finger
• spelling
• webcam
• software
• eepis
• its
• teknik
• informatika

[nakagawa53]

Sorry about my (very) late reply. There are many people asking the same question.

Please note that you cannot extract the program, compile it, and hoping that it would work right away =_=" there are several libraries that you need to install.

I have updated the info section I hope it helps.

[baamchika]

How do you run this program? :)

[nakagawa53]

@baamchika Info section updated.

[MrSimplesify]

waw... keren bgt... saya juga lagi berusaha untuk hand detection in c# juga ^^.. btw saya sudah kerem message ke @nakagawa53 .. mohon di baca ya ^^.. hope a great response

[nakagawa53]

@MrSimplesify Info section barusan saya update boss

[ank3201]

hey...

great work.

i visited the link but was unable to run the code. it wasn't detecting my hand. can u give me some solution how i can go about doing it

[nakagawa53]

@ank3201

First of all make sure that the haar training detection works. If it doesn't maybe you should consider making your own haar training. The haar cascade I used works well in well-lighted roomt. If the haar detection works, then the hand shape classification should work as well.

Please note that I used 640x480 resolution with Logitech Quickcam 9000.

[rajpriy1629]

Sir are u got ur answer , if yes plz share it i got the same problem

[maanbustamante]

@nakagawa53 tried running your program but got this error --> src_image = highgui.CvLoadImage(file, highgui.CV_LOAD_IMAGE_GRAYSCAL­E);

[maanbustamante]

@nakagawa53 libraries such as? have you completed the 7 other signs left? thanks!

[virtualboy50]

hello @nakagawa53.

greet work, i downloaded code & libraries ( it's give a run time error on KNearest.cs

"src_image = highgui.CvLoadImage(file, highgui.CV_LOAD_IMAGE_GRAYSCAL­E);"

i think it's need a the training files.

would you share them please, or tell me what to do?

thank you so much

i noticed that there was

link for that but i missed now

[adi2kx]

Hey @nakagawa53. Have couple of questions. i) Does it detect hand when u bring the hand ahead of the face ? ii) Does using Haar clasifiers slow down the program ??

[DexteriousDex]

Thanks for the noob guide :D