Hand Gesture Recognition Based On Motion Sensor

Human-computer interaction as a part of artificial intelligence is becoming the focus of concern. It is about how to communicate with the machines. The keyboard, touch-screen, voice recognition was the mostly used traditional way. But they have lots of flaws, which made people begin to look for more efficient and natural way. The purpose of this paper is to design a 3D dynamic hand gesture recognition systems.More mature technology now is based on machine vision. But for the limitation of camera, it is hard to get promoted. So the input in this paper used motion sensor to get the movement information, including acceleration, angular velocity and magnetic induction intensity. We solved the attitude angles through the quaternion method. More commonly used recognition algorithm is hidden markov model and template matching, etc.. But due to the reason that the set of gestures in this article has a certain degree of differentiation, so "characteristic analysis method" is proposed. This paper identified six types hand gestures: mobile, rap, rotating and shaking, check and drawing fork, which rotating and mobile also need to be further identified the specific direction, and selected five characteristics:length, energy, wave number, the unilateral angular velocity, the largest shaft of angular velocity energy, to establish classifier, first identifying the gesture categories, then the specific direction.In order to verify the validity of the algorithm and estimate its accuracy, this paper also built a wearable hardware test platform, to get the real-time gesture movement data and did the data filtering and the attitude algorithm. Then we transfered these data to PC for processing characteristics analysis algorithm. The average precision reached 89.2%, which guaranteed the accuracy and efficiency.