Research And Implementation Of Dynamic Hand Gesture Based On Parallel HMM

The recognition of dynamic hand gesture using wireless signal always exists the problem of high computational complexity and long recognition time.Through the analysis of wireless frame structure,the preamble's signal of 802.11a is collected through Software Defined Radio platform and reserved as the data source.In addition,more than one time-domain feature sequences perform unique shape for different dynamic hand gestures.This paper proposes two models for dynamic hand gesture classification based on the time-frequency features and channel characteristics of preamble.On the one hand,the received signals are split into single-cycle and the unavoidable electronic interference is reduced through Discrete Wavelet Transform.At the same time,due to fuzziness of dynamic hand gesture,the amplitude and duration for the same dynamic hand gesture are not exactly same.Therefore the parallel HMM models which represent for different hand gestures and features are built for recognition.The result shows that the average recognition rate is about 90.5%for dynamic hand gesture recognition.On the other hand,the qualitative relation between the received signals and the hand motion feature(gesture motion direction,range of motion,motion cycle and movement speed)are directly built with a real-time recognition system.And finally the average accuracy of dynamic gesture recognition can reach 97%.