Dynamic Gesture Recognition Research Under Human Basic Motion

With the advancement of technology, the mobile electronic devices equipped with a wealth of hardware facilities such as MEMS inertial sensors, which not only make people’s lives more convenient, but also provide a good platform for new human-computer interactions such as voice recognition, image recognition, gesture recognition and so on. Image recognition is vulnerable to the effects of light and voice recognition is susceptible to noise interference. Thus, gesture interaction which is based on the inertial sensors to interact with its unique advantages, become a research hotspot. Dynamic gesture recognition when the human body is stationary has made great progress. Many scholars have tried different recognition methods and verify its validity. However, at this stage, there is almost no paper which studies dynamic gesture recognition when human body is in motion. This paper will start in this direction, research on dynamic gesture recognition under the condition of human movement.By using inertial measurement unit Sensor-Box Nokia provides, this paper mainly focuses on dynamic gesture recognition under the condition of human basic movement such as walking,walking up and down stairs, taking the elevator and so on. In this paper, the basis human movement was classified and this paper analyzed the human basic movement’s influence on the dynamic hand gestures. Combined research objectives, the paper proposed three identification schemes for dynamic gesture recognition under conditions of human movement.Based on eliminating human basic movement interference, this paper proposes an identification scheme based on dual inertial sensor method and an identification scheme based on mathematical model method. From the perspective of feature classification recognition, this paper proposes an identification scheme based on the threshold increasing recognition program.In order to reduce the error of inertial sensor, this article established the Sensor-Box accelerometers and gyroscopes error model. And use six-position method to calibrate the deterministic error. ARMA model of acceleration random error signal is established with the method of time series analysis and the acceleration random errors classical were effectively filtered out with Kalman filter. For the gyro random error signal, using the standard Allan variance analysis to identify its main random error term, and using wavelet analysis to separate its noise signal effectively.Dual inertial sensor method is that acquisition of motion signal and human dynamic hand signals simultaneously by both Sensor-Box and then to eliminate the interference of human basis movement based on the relative motion theory. The test results of dual inertial sensor method meet the requirements. Mathematical model method is that different mathematical models are proposed for different types of foundation movement and the model construction methods are given. For periodic signal such as human walking up and down stairs etc, this paper proposes a method to establish the mathematical model of a periodic signal. Test results of mathematical model method meet the requirements and the algorithm still needs to improve. Increasing the threshold method is based on the feature classification method. Find out the classification characteristics of dynamic hand gesture signal (flip, swinging, shaking etc) and build a classifier. Test result of increasing the threshold method is good.