Research On Gesture Control Systerm For Autonomous Intelligent Humanoid Robot

Intelligent robot have the ability to perceive external information for havingseries of sensors. According to the way that intelligent robot deal with theinformation, we can devided them into semi-autonomous intelligent robot andfully-autonomous intelligent robot, the semi-autonomous intelligent robot send thedigital information which are collected by the communication module to PC, andthen the computer complete the information processing and send the result back tothe robot, the robot don’t have the ability to deal with the information by itself. Butthe fully-autonomous intelligent robot can compute with the information all by itself,it has a real similarway with the human.In order to improve the naturalnes between human-robotinteraction, we shouldtry to make the robot learn the form of communication of human, and make therobot can communicate with human by using human communication.This systemuses vision-based gesture recognition technology as a way of how robot understandhuman gestures. The Experimental platform is a fully autonomous humanoidintelligent robot which is research and development by mobile computing center ofShenzhen Graduate School of Harbin Institute of Technology.This paper describes the development status of the vision-based gesturerecognition technology, discussed the basic theory of the related technologies, thencombined with the processing performance and technical characteristics of the fullyautonomous humanoid intelligent robot, designed a real-time gesture controlsystem.In this issue, since the fully autonomous intelligent robot have real-timerequirements and its own processing performance are limitated, how to maintain ahigh recognition rate in a real-time system is the focus of our research.In the section of skin-color extraction, we build a mixture Gaussian Modelbased on the index of the luminance in YCbCrspace, and after the image denoisingand the extraction of the maximal connected domain obtained we get the wholeclear guesture picture.In the part of feature extraction, the systerm use distancetransform in the hand center extrationalgorithm, designed a curvature-basedfingertip detection algorithm, then designed a classification method by use thefingertip characteristics and the guesture gaussian model. Finnally, the average recognition rate reached88.22%.