============================= Trainable Gesture-Recognition based on Hidden Markov Models using JavaScript, WebGL and WebRTC.
System-Requirements:
- Latest Google Chrome
- WebGL and WebRTC support
- App running on a server (cross-origin-policy)
- Connected webcam
-
Training (train.html) ======================== To train the HMMs, videos of sample gestures are used (videos/**/tXX.m4v). For the prototype the corresponding videos are located in the videos folder. The result is a HMM representing the equivalent gesture. The training configuration is found in the "js/train.js" file.
-
Recognition (recognize.html) =============================== The recognition process is based on the trained models which are configured in "js/models.js". To calibrate your skin tone hover your hand in front of your webcam filling the rectangular area in the live preview. A click on the canvas of the live preview recalculates the HSV values of your skin. Better results are achived under different lighting conditions.
-
Validation (validate.html) ============================= The validation process uses independent validation samples and calculates matches between the underlying models and validation videos (videos/**/vXX.m4v).
- Author: Adrien Zdanski
- Github Repo: http://github.com/adrienzdanski/js-hmm-gesture-recognition