A Hand Gesture Recognition System Based on RGB-D Sensor

• Main Authors: Wen-Cheng Hsu, 許文政
• Other Authors: 作者未提供
• Format: Others
• Language: zh-TW
• Published: 2015
• Online Access: http://ndltd.ncl.edu.tw/handle/69261574935610420626

碩士 === 銘傳大學 === 電腦與通訊工程學系碩士班 === 103 === Hand gesture control is mainly a human-machine interactive input device developed by using machine vision. It extracts the operator’s gesture video with camera and cuts the hand area through graphical processing. It then extracts the valid characteristics to present the hand gestures. Finally the output results can be obtained through classifier. To achieve the light, small and low-cost purposes, the demands for realizing these key algorithms in the embedded system with limited resources are getting higher, which however increases the overall complexity of the solution. This study aims to develop a robust hand gesture recognition system. We firstly calibrate the operator’s RGB video and depth video extracted by Kinect sensor, so as to synchronize the video signals. Then we filter the complicated background information other than the arm, so the skin color model can be used to position the operator’s hand area correctly. After finishing the pre video processing, we conduct cluster analysis with the profile characteristics of the horizontal project for the depth video of the hand gesture. We also classify all training data into multiple types based on the principle of no wrong classification. After the classification, we then make up the characteristic vector of each hand gesture with the bag-of-words model composed of SIFT descriptors. After that, we put the hand gesture characteristics of the same type into the corresponding SVM-based classifier for supervisory learning, so as to build the multi-expert classification system with high recognition. The experimental results show, the method proposed in this study can indeed recognize the operator’s hand gesture effectively.