A Novel Wearable Emg-based Text-entry System

Wearable smart devices,such as smart watches,have been rapidly developed in recent years.However,due to the small screen and other reasons,the input of these smart wearable devices is greatly limited,so the interaction issue has become a key factor restricting the development of wearable technology.At the same time,social development has also led people to pay more and more attention to personal health,making smart devices that monitor physical information also develop,not just heart rate detection and other functions.Electromyography(EMG)can also be used for human-computer interaction.In the field,such as gesture recognition.However,the current research methods can only identify large actions,or only a small number of actions,or require multiple hardware support.Under this background,the subject designs a wearable text input system based on the myoelectric signal to solve the input constrained problem,and can use fewer hardware devices to realize fine-grained keystroke recognition and achieve the effect of a virtual keyboard.In order to realize the function of signal acquisition and processing,first of all,this subject has used commercial hardware accessories to design a set of intelligent wristbands for collecting the electromyographic signals generated by the muscles of the forearm when the finger taps the keyboard.Then,in order to process the collected original signal,an EMG signal with low loss and high signal-to-noise ratio is obtained.This project has designed an effective noise suppression algorithm for the system.Then,this topic proposes the definition formula of the change amount,and designs an algorithm based on the change amount to detect and extract the myoelectric signal(action signal)generated by the action.In addition to this,the design algorithm of the subject analyzes the motion signal and implements the text input function.First of all,this topic has designed an embedded feature engineering for selecting an effective feature set.Then,the system establishes a model based on a variety of classification algorithms,and realizes the functions from the feature calculation of the EMG signal to the character recognition and evaluation.Finally,based on the results of character recognition,this paper redesigned the text correction model to correct the word,so that the system can output the text content more accurately.Finally,this project evaluates the system through a large number of experiments under different experimental scenarios.It is found that the system can achieve high character recognition accuracy of 87.5% and 89.5,respectively,regardless of whether characters are input on a physical keyboard or a paper keyboard.%.Different experimenters use this system,the character output accuracy of up to 95.1%,the lowest is 82.9%.For word errors due to character recognition errors,the system uses an error correction model to correct it.Compared to uncorrected results,the output word accuracy can be increased by 57% to 93%.