A real-time gesture recognizer based on dynamic programming

Abstract

Coded gestures are physical movements to which precise meanings have been attached. Their use with an electronic pointing device potentially increases the power of the pointer by allowing several distinct control functions to be incorporated. The formation of a coded gesture is invariably subject to distortions which make its automatic recognition difficult. After particular consideration of three techniques, a dynamic programming algorithm was developed and was used in the design of a real-time gesture recognizer that allows mouse-driven software to be operated purely by means of an electronic pointer. The decision was taken to implement the gesture recognizer using a widely available, obsolescent microcomputer in order to keep down costs. Despite the inevitable technical compromises in this approach, results of preliminary trials using 10 normal subjects show an average gesture recognition rate of 98.3% and a gesture confusion rate of only 0.4%.