Hand Gesture Recognition using a Modification of the Growing Neural Gas Algorithm

Classification: Files related to classification module.

• classifier.cpp
• gesture.cpp
• gesture.h
• gestureexamples.cpp
• gestureexamples.h

Segmentation: Files related to the segmentation module.

• haarcascade_frontalface_alt.xmlhandDetector.h
• handDetector.h (range version)
• handDetectorHist.h (histogram version, selection is a square)
• handDetectorHist2.h (histogram version, selection is more precise)

Learning: File that use the segmentation and extraction module to extract features from sample gesture image and store them in csv files.

• learning.cpp

Example application: Launch the camera so that you can see the gng-t graph on your hand. (use all the module except the classification one)

• main.cpp

GNG-T: Files related to the features extraction module.

• gngt.h
• tracking.cpp
• tracking.h
• unionfind.h
• param.txt
• param_fast.txt

You can compile the different program using:

• make video for the Example application
• make classifier for the classification module
• make recorder for the recoreder program
• make learningfor the learning pogram

make clean erase the executables.

We use the Boost graph library as well as opencv so you will need those library to be able to compile.

After compilation there are multiple ways to run the above program:

1. ./video

   Two windows appear,'track' and 'skin'. In the 'track' window, perform the gesture, select a region of your skin(alternatively the code can also detect skin automatically), and press q.

   ./classifier ToBeClassified.csv features1.csv features2.csv features3.csv features4.csv features5.csv features6.csv features7.csv features8.csv features9.csv

   This will give the output as which feature(predefined gesture), the gesture you performed most closely resembles.

2. ./classifier Example.csv features1.csv features2.csv features3.csv features4.csv features5.csv features6.csv features7.csv features8.csv features9.csv

   The Example.csv already contains the features of a number of gestures that were performed by us, and the above command gives the output as the feature(predefined gesture) that most closely resembles the gestures in the Examples.csv file.

3. ./recorders/recorder ./recorders/Trial

   A window will appear called 'img'. Perform the various gestures you wish to identify and press 's' to select one frame for every gesture. Press q when you have completed all the gestures.

   ./learning recorders/Trial

   ./classifier recorders/Trail/features.csv features1.csv features2.csv features3.csv features4.csv features5.csv features6.csv features7.csv features8.csv features9.csv

   This will give the output as which features(predefined gestures), the gestures you performed most closely resemble.