Real-Time Gesture Recognition by Means of Hybrid Recognizers

  • Authors:
  • Andrea Corradini

  • Affiliations:
  • -

  • Venue:
  • GW '01 Revised Papers from the International Gesture Workshop on Gesture and Sign Languages in Human-Computer Interaction
  • Year:
  • 2001

Quantified Score

Hi-index 0.00

Visualization

Abstract

In recent times, there have been significant efforts to develop intelligent and natural interfaces for interaction between human users and computer systems by means of a variety of modes of information (visual, audio, pen, etc.). These modes can be used either individually or in combination with other modes. One of the most promising interaction modes for these interfaces is the human user's natural gesture.In this work, we apply computer vision techniques to analyze real-time video streams of a user's freehand gestures from a predefined vocabulary. We propose the use of a set of hybrid recognizers where each of them accounts for one single gesture and consists of one hidden Markov model (HMM) whose state emission probabilities are computed by partially recurrent artificial neural networks (ANN).The underlying idea is to take advantage of the strengths of ANNs to capture the nonlinear local dependencies of a gesture, while handling its temporal structure within the HMM formalism. The recognition engine's accuracy outperforms that of HMM- and ANN-based recognizers used individually.