Research On Automatic Recognition Of Basic Strokes Of Handwritten Chinese Characters Based On FMCW Radar

In recent years,with the development of computer technology and human computer interaction(HCI),it has brought qualitative changes to everyone’s life and also added new design concepts to the field of HCI.However,most traditional input technologies heavily rely on touch-contact devices and non-contact computer vision to obtain character information,which limits further applications in bacterial infection environments,ultrahigh/low temperature environments,and light-affected scenarios.This thesis proposes a non-contact automatic recognition scheme for basic strokes of handwritten Chinese characters based on frequency modulated continuous wave(FMCW)radar and carries out corresponding work.The main results of this thesis are as follows:(1)A platform for collecting the original data of basic strokes of handwritten Chinese characters is designed.The combined module of IWR1443 and DCA1000 hardware is used to collect the data on eight strokes of dot,horizontal,lift,left falling,bend,right falling,vertical and hook after the characteristic word " 永 " is decomposed,and the corresponding basic stroke data set is constructed.(2)A method of extracting parameter feature information from raw data is studied.Aiming at the interference of the experimental environment and spectrum leakage,the parameter configuration of low pass window filter Fourier transform(LPF-Window-FFT)algorithm is studied,and nd the distance estimation is performed on the original data to obtain the range time sequence(RTS),and secondly,in order to solve the problem of angle time series and insufficient of super resolution array estimation,the frequency domain Capon(FD-Capon)algorithm is proposed to estimate the azimuth of the original data to obtain the azimuth time sequence(ATS).(3)A FA-FBO algorithm is proposed to make the extracted parameter features more prominent.Aiming at the noise and other interference problems existing in RTS and ATS images,a series of image processing operations such as feature area framing,binarization,and opening operations are carried out to solve the partial interference of the image except for the features.The range time sequence feature map(RTSFM)and the azimuth time sequence feature map(ATSFM)are obtained.(4)The work of extracting the feature information of the basic stroke feature atlas of handwritten Chinese characters is carried out.This paper designs a convolutional neural network(CNN)model to extract the shallow spatiotemporal features of RTSFM and ATSFM,and uses the feature map set to train the network model.The decomposed eight basic strokes achieve an average recognition accuracy of 99.25%.