Real-time vision-based telepresence robot hand control

With the rapid development of computer vision, the Internet and robot technology, it´s potential to develop a telepresence robot for providing service for people who need a real avatar. We present a real-time vision-based telepresence robot hand control system, including the hand detection, hand tracking, hand recognition, 3D hand position and hand control modules. From our experiments, the composition of the Adaboost detection framework, random forests based skin model and the hand tracking method, can detect the hand in the video stream with the accuracy about 91%. Hand recognition incorporating the linear dimensionality reduction and RBFSVM, can attain the accuracy about 97%. Our proposed approach achieved a good performance even in some extreme cases. Combined with the well-designed communication protocol, we accomplish the goal of interacting between human and the robot naturally and efficiently.