Publication: Camera-based gesture recognition
Date
2021-07-01
Authors
Jeanne Leong, Yee Jing
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Hand gesture recognition has been used in many applications. However, previous work is to research what application has been made, which method will give the best performance on gesture recognition and what method to train algorithm will give the best accuracy on gesture recognition. This thesis aims to investigate the algorithm for gesture recognition, decode the actions and hand gestures into standard JavaScript Object Notation (JSON) format and develop an application for gesture recognition. Therefore, data classification is used in this project to recognize the hand gesture. Firstly, hand detection is used to detect the appearance of the hand. Then, feature extraction and classification are done by using Convolutional Neural Network (CNN). Data of the images are obtained and trained. There are two experiments carried out to test the accuracy. The accuracy of the first experiment is 96.8% when the white background is used for prediction while the accuracy of the second experiment is 66.6% when the complex background is used for prediction. The decoded action and hand gestures is developed into a web-based application. 5 simple gestures for web-based applications are made. The web-based application can be used to control web browser. This web-based application can be further extended for other functions like games, menu settings for televisions and others in the future.