REAL-TIME CLASSIFICATION OF CONTINUOUS GESTURES

This study aims to develop a real-time continuous gestures classification system. The approach does not rely on using any gloves, visual markings, or special camera to achieve the recognition task. Instead, it uses just real-time video streaming of bare hand from a webcam. The top priority of this work is to create a robust, accessible, and flexible system that capable of dealing with different kinds of gestures (those that are solely characterized by the hand shape but not the motion, and those that required a specific hand or fingers movement) and determining how and when the system should respond to each gesture. The framework for handling both gesture types is a novel approach, and the evaluation shows promising results at 91.77% average correct classification using all 26 American Sign Language (ASL) manual alphabets as the examples. The system also achieves real-time performance at an average of 25 FPS. Other novel approaches in this study include the use of adaptive background model with the combination of elliptical skin-color model, K-mean clustering, and local region Gaussian mixture to detect hand under complex background. The configuration that allows the system to respond promptly to the gesture when it starts or ends, and the use of fingers counting to help increase the accuracy during the detection and classification stages are also developed.

Name thesis:degree_name

Doctor of Philosophy

Level thesis:degree_level

Campus Only Dissertation

Discipline thesis:degree_discipline

Electrical and Computer Engineering

Year

2015

Author dc:creator

• Maitree, Rapeepan

Contributors dc:contributor

• Gupta, Lalit

Repository record dc:identifier

https://opensiuc.lib.siu.edu/dissertations/1007

OAI identifier oai:identifier

oai:opensiuc.lib.siu.edu:dissertations-2011