AlexNet Optimization And Bimodal Pen-holding Gesture Recognition

Due to the irregular pen-holding gesture of current primary and middle school students,problems such as poor writing quality,poor sitting posture,myopia in the eyes,and hyperplasia of the finger joints have caused adverse effects on their physical and mental health.In response to the problem of pen-holding gesture,Chinese researchers have conducted more theoretical research on it.The country has also issued "Several Opinions on Strengthening the Teaching of Writing in Primary and Secondary Schools" and other documents,which not only emphasize the importance of correct gesture for writing by students,and clearly define the pen-holding gesture for primary and middle school students.However,the current pen-holding gesture and calligraphy teaching mainly use face-to-face and video teaching,lacking of research and practice of automated teaching feedback systems,and there is no public data set for pen-holding gesture.Therefore,based on the project,this paper studies the construction of pen-holding gesture data set and the automatic recognition method of pen-holding gesture.The main research contents are as follows:(1)In order to solve the problem that there is no data set of pen-holding gesture,this paper,on the basis of referring to various literatures related to pen-holding gesture,classifies pen-holding gesture into nine categories,including standard type,misplaced type,horizontal type,counter-head type,twisted type,torsion type,fist type,sleeping type and straight type.The data set of pen-holding gesture is shot respectively against the background of white paper,pigment-like desktop,white desktop,messy desktop,pure black background and notebooks full of writing.900 of each gesture are taken,a total of 8,100.(2)In order to solve the problems of AlexNet's insufficient feature extraction of pen-holding gesture,slowing convergence speed of network and lowing recognition rate of pen-holding gesture,this paper proposes a method of AlexNet optimization and bimodal recognition of pen-holding gesture.In this paper,batch normalization and attention mechanism are used to optimize the AlexNet network structure,and based on the optimized AlexNet network,the bimodal gesture recognition of pen-holding is realized.Firstly,the gesture image is segmented and the skeleton image is extracted to obtain the bimodal images.Then there are pretreated;Finally,the processed pen-holding gesture images are input into the optimized AlexNet for feature extraction and feature fusion,then Softmax classifier is used to recognize the pen-holding gesture.(3)On the basis of the above research,this paper designs and implements the prototype of penholding gesture recognition system.This system consists of four parts: pen-holding gesture image reading,bimodal images acquisition,bimodal images preprocessing and pen-holding gesture recognition.According to the characteristics of pen-holding gesture,this paper builds nine types of penholding gesture data set,and proposes a bimodal pen-holding gesture recognition method which is based on optimized AlexNet.On the basis of this research method,a prototype of the pen-holding gesture recognition system is designed and implemented.The pen-holding gesture recognition in human-computer interaction has carried out preliminary practical exploration.