Research On Recognition Technology Of Volley Handwritten Inkless Chinese Characters Based On Deep Learning

Chinese people’s daily life and work are inseparable from Chinese character writing and recognition.At present,most of the commonly used writing methods are contact writing.However,with the continuous development of science and technology,people hope to have a more intelligent,free and environmental friendly writing method,which can be written by hand and in the air.At present,in the field of target detection and tracking and handwritten Chinese character recognition,more people begin to use deep learning models with higher accuracy and faster speed.Based on the consideration of simple and easily available acquisition equipment,this paper uses the most common single RGB camera,and uses deep learning models such as Yolo V4 and improved deep convolution neural network,This paper presents a recognition method of airborne handwritten and inkless Chinese characters based on deep learningFirstly,in the aerial handwriting part,this paper makes a fingertip data set containing 15264 hand images,and then obtains the fingertip detection model based on Yolo V4 through training,which is connected end-to-end with the fingertip tracking algorithm.Secondly,this paper proposes an improved fingertip tracking algorithm based on Kalman filter:the intersection and union ratio evaluation matrix of fingertip detection frame and fingertip prediction frame is used to match it,which effectively reduces the number of fingertip ID conversion and solves the problem that the fingertip cannot be tracked when the fingertip is temporarily blocked to a certain extent;When the fingertip detection model can not detect the fingertip,the K curvature method is used for secondary detection and fingertip correction to improve the accuracy and continuity of tracking.The improved fingertip tracking algorithm can effectively complete the tracking when multiple fingertips,fingertips disappear temporarily or cannot be detected,and obtain the tracking trajectory.Compared with the Kalman filter tracking algorithm,the tracking accuracy is improved by 11.2%,which can better complete the work of Chinese character handwriting in the air.In addition,in the part of Chinese character recognition,this paper selects three data sets for training,and then proposes an improved deep convolution neural network based on lenet-5t:using sparse input layer,the computational complexity and computing time of training are reduced;After the input layer,a feature enhancement layer is added to provide a priori knowledge by using path integral feature and eight direction feature;Fewer and smaller convolution kernels and more convolution layers are used to effectively retain more spatial information;A maximum pool layer is connected behind each convolution layer for downsampling,which constructs a deeper neural network and improves the operation speed;The overall network uses leaky relus activation function and adds softmax function in the output layer to complete the classification of handwritten Chinese characters.The improved network effectively improves the recognition accuracy of handwritten Chinese characters.