Hand Gesture Recognition Based on Dynamic Bayesian Network

• Main Authors: Kai-JungChung, 鍾凱融
• Other Authors: Jenn-Jier Lien
• Format: Others
• Language: en_US
• Published: 2013
• Online Access: http://ndltd.ncl.edu.tw/handle/55769250383578545818

碩士 === 國立成功大學 === 資訊工程學系碩博士班 === 101 === In this thesis, we construct a hand gesture recognition system based on dynamic Bayesian network model through using the human skeleton information captured by Kinect sensor. We estimate the model parameters of the dynamic Bayesian network with Expectation-maximization algorithm and the features including motion direction of both hands and the relative position between both hands and the face, and trained a gesture recognition system which is suitable for both one-hand and two-hand gestures. In the experiments, we focus on 10 common hand gestures of volleyball referee. We trained the hand gesture recognition system with the 1000 video sequences which be collected by Kinect sensor and tested the trained DBN models of hand gestures recognition system with cross-validation. Finally, we analyze and compare the recognition models according to the experimental results, and explain that accuracy of the hand gesture recognition system.