Skip to content

Latest commit

 

History

History
executable file
·
79 lines (42 loc) · 1.9 KB

README.md

File metadata and controls

executable file
·
79 lines (42 loc) · 1.9 KB
Raw

LeapMotionGestures

A MATLAB Toolbox that performs Gesture Recognition using the Leap Motion signals.
The demo example is adapted to a Digit Recognition Application. It can classify an input sequence into a type of digit 0-9.

This repo contains:

  • A small database of leap motion recorded digits
  • A preprocessing module
  • A feature extraction module
  • A HMM based sequence classifier
  • A GMM based sequence classifier
  • A DTW based sequence classifier
  • A fusion rule module

Preprocessing

To normalize data and ease recognition, we apply the following preprocessing pipeline

Sequence Classification

It allows training 3 different types of models for digit recognition:

  • Hidden Markov Models (HMM)
  • Gaussian Mixture Models (GMM)
  • Dynamic Time Wraping (DTW)

Hidden Markov Models (HMM)

Gaussian Mixture Models (GMM)

Dynamic Time Wraping (DTW) + KNN

Fusion Rule

Models outputs are combined to improve decisions

External Dependencies

The repo has the following dependencies:

DTW Library: http://www.mathworks.com/matlabcentral/fileexchange/43156-dynamic-time-warping--dtw-

HMM Library: https://www.cs.ubc.ca/~murphyk/Software/HMM/hmm.html