Research On Handwriting Recognition Method Based On Sound Signal

In the intelligent era,the interaction between people and various intelligent devices become more and more frequently,and the development of intelligent perception technology has also brought various contact and non-contact devices into people’s lives,which has a huge impact on the way of human-computer interaction.In a variety of human-computer interactions,writing recognition is a natural and friendly way.By tracking the user’s writing behavior and recording the writing trajectory,the trajectory information can be converted into computer code.At present,there are many ways to realize handwriting recognition.(1)Handwriting recognition based on wearable devices collects writing action information and completes recognition tasks by allowing users to wear sensor devices.This method provides rich and powerful hardware capabilities,but wearable devices are often costly,and additional hardware makes users feel inconvenient.(2)Image-based writing recognition uses a camera to capture the image of the user when writing,and then uses image processing technology to detect and recognize the user’s writing.This method can obtain a high recognition rate,but has disadvantages such as high hardware cost,difficult deployment,and high requirements for conditions such as illumination and occlusion.(3)Wi-Fi signal-based writing recognition uses the channel state information in the Wi-Fi signal to perform writing recognition.The implementation cost of this method is low,but because Wi-Fi signals are not universal in people’s lives,the universality is poor.In response to the problems in the current handwriting recognition field,we propose a handwriting recognition method based on sound signals.Compared with other methods,this method has the advantages of low computing resource requirements,no additional auxiliary equipment,and good user experience.The research content of this article mainly includes the following points.(1)Using the microphone and speaker components that are ubiquitous in smart devices,the fine-grained tracking of the hand’s trajectory is achieved through downconversion,distance measurement,and multipath elimination,so as to Obtain information about the distance change of the hand relative to the smart device.(2)Use filtering,derivation and other methods to preprocess the data,and then design a motion detection method for sound time series data based on the idea of subwindow merging.Use the energy difference between the motion signal and the noise signal to obtain the start and end points of the writing motion,and finally to achieve a better segmentation effect,a series of optimization strategies are proposed to achieve precise segmentation of writing actions.(3)Convene volunteers and use smart phones to collect the written information of 26 uppercase English letters;design experiments to compare the classification effect of five models including support vector machines,random forests,K nearest neighbors,deep neural networks,and convolutional neural networks,and then select the best classification model;design experiments to analyze the impact of data scale,data enhancement methods,and letter categories on the classification effect of the model.Finally,this paper has determined the best classification model based on the convolutional neural network through experiments.The model has achieved 81%,81%,80% in the recognition accuracy,recall rate and F1 value of 26 capital English letters,which proves the feasibility of using sound signals for writing recognition.The research in this article provides new ideas for writing recognition methods based on sound signals,and also provides a reference for future exploration of new human-computer interaction methods,and also contributes to people’s pursuit of more convenient and better life.It has certain application value and social significance.