Design Of Virtual Human-machine Input System Based On MEMS Gyroscope

With the continuous progress of science and technology,wearable devices have been used in a new way in the field of human-computer interaction.The research of smart wearable devices based on MEMS sensors has become an emerging direction.As a new input method,the use of wearable devices has become more diverse.At present,the input method of wearable smart devices is to let users make corresponding actions to complete the input according to the set gestures,which increases the threshold and time cost of users.The purpose of this paper is to design a virtual human-computer input system,using small size and light weight MEMS sensors to make a wearable device,which learns and recognizes the finger gestures when typing by establishing a recurrent neural network,so that users can get rid of the physical keyboard after wearing it,and simply make tapping movements on the plane according to their own input habits to complete input.The main research contents of this paper are as follows.(1)The hardware design of the virtual human-computer input system includes MEMS sensor,data acquisition board and host computer.the MEMS sensor integrates gyroscope,accelerometer and magnetometer,which can detect acceleration,angular velocity and other data in real time in the three-axis direction.By fixing the MEMS sensor on the front of each finger,the posture data of each finger can be collected while typing.The data acquisition board with STM32F1 chip can collect and process the MEMS sensor data on each finger in real time and send it to the host computer,which will recognize the data and complete the output after receiving it.(2)MCU fuses the data of MEMS gyroscope,accelerometer data and magnetometer data to reduce the data drift of MEMS gyroscope and sends the raw data to be decoded into easy to understand acceleration,angle and other information.The upper computer preprocesses the received data such as normalization to produce a data set for training the recognition model.(3)For the specific gesture recognition input of wearable devices currently available in the market,this system builds a Long Short Term Memory(LSTM)network model for gesture recognition.It learns the user’s input habits to improve the universality of the system.the MEMS gyroscope data will drift with time due to the error,and its data is standard time series data.the LSTM can effectively solve the long-time dependence problem through its special gating structure,so as to suppress the gyroscope drift twice and improve the recognition accuracy of the system.(4)The studied input system is equipped with a probability-based error correction function.It will check and correct errors after each completed word input to further improve the recognition accuracy of the system.Before using the MEMS gyroscope based virtual human-machine input system designed in this paper,users are required to use the physical keyboard to tap the alphabet keys for typing training according to the standard typing fingering;when using,users only need to make typing actions in the plane according to the previous typing posture,and the trained neural network model is called to recognize the gestures and output the corresponding key values.The test results show that at a typing speed of 120 English letters per minute,the recognition algorithm can reach 95% correct rate,and after corrected by the error correction algorithm,the accuracy rate is improved by 97.8% to meet the normal typing needs.