Research On Hand Gesture Detection And Recognition Method Based On FMCW Radar

With the development of the field of artificial intelligence,the Human Computer Interaction(HCI)has gradually become a hot research topic.Hand gesture recognition,as an effective way of HCI,has also been widely concernd.Among the existing hand gesture recognition devices,the radar stands out among many devices because of its strong antijamming ability,low power consumption and privacy protection.However,there are still some problems in the research of radar-based hand gesture recognition,such as low recognition accuracy and complex data input.To solve these problems,this thesis carried out the following research based on Frequency Modulated Continuous Wave(FMCW)radar.1.There is a lack of research on continuous hand gesture.In this thesis,a continuous hand gesture segmentation method based on time sliding window detection is proposed.Firstly,the radar continuous hand gestures data are collected,which is analyzed in time domain and frequency domain respectively according to the parameter estimation results of thee continuous hand gestures,so as to segment the continuous hand gestures accurately.Finally,this thesis designs the hand gesture detection range and evaluates the segmentation accuracy of the hand gesture.2.In view of the lack of dimension of radar hand gesture information,this thesis studies the construction method of multi-dimensional parameter data set.Firstly,the RangeTime Map of hand gesture is obtained based on Fourier Transform.Secondly,the hand gesture Doppler-Time Map is obtained based on the 2-Dimensional Fourier Transform.Finally,the hand gesture Angle-Time Map is obtained based on the Multiple Signal Classification.In this thesis,the three hand gesture dimensions are used to fully extract the hand gesture features,which provides a data basis for the multi-dimensional parameter hand gesture recognition research.3.To solve the problems of low efficiency of hand gesture recognition and complex hand gesture training model,this thesis proposes a Fusion Dynamic Time Warping(FDTW)algorithm based on clustering.Firstly,the hand gesture Map with parameter estimation is used for clustering operation,and the Central Trajectory Map(CTM)obtained improves the operational efficiency of hand gesture recognition.Secondly,this thesis improves the traditional Dynamic Time Warping(DTW)algorithm,and uses the three hand gesture feature information after clustering as the three input branches of FDTW.Finally,the hand gesture recognition was accomplished by judging the fusion decision distance.