Design And Implementation Of Physiological Monitoring System Based On Wireless Headsets

Wearable physiological monitoring system uses wearable biosensors to collect human movement and physiological parameters to realize the management of wearer’s movement and health.With the increasing number of the elderly population,sub-health groups and patients with chronic diseases in China,people’s awareness of health management has gradually improved,and higher requirements are put forward for wearable physiological monitoring system.The traditional wearable physiological monitoring system usually collects a single physiological signal at the wrist.In the daily living environment,users often forget to wear it,and this kind of system usually does not fully tap the connotation of sports and physiological signals.In addition to essential entertainment features,wireless headsets have the advantages of very little psychological pressure and more accurate acquisition of motion and physiological signals.This thesis designs and implements a physiological monitoring system based on wireless Headsets by integrating the technologies of Internet of things,cloud computing and deep learning.The main research contents and work of the thesis are as follows:Firstly,a software and hardware system that can collect sports and physiological signals is based on wireless headsets.The system is divided into three parts: the headset end,the mobile end and the cloud server.In addition to the entertainment function,the headset end is also responsible for the collection of motion and physiological signals;the mobile end is mainly responsible for the control of the headset,the visual display of physiological information,and the algorithm service of calling the cloud host;the cloud server is used for fall detection and physiological monitoring algorithms It is deployed and provides service interfaces.If an abnormal event occurs,the guardian can be notified remotely or the hospital can be called for rescue.Secondly,a fall detection algorithm based on head inertial sensor is designed and implemented.Using a combination of complementary filtering and extended Kalman filtering to filter the original acceleration and angular velocity,reduce the drift error of the gyroscope and filter the Gaussian white noise of the acceleration to obtain more accurate motion data;For the problem of different reference systems,quaternion coordinate transformation is used to map the headsets coordinate system to the geodetic coordinate system;for the disadvantage that the threshold method cannot distinguish multiple types of falls,a multi-channel one-dimensional convolutional neural network(Mutil-Channel 1DCNN,MC-1D-CNN)is proposed.MC-1D-CNN fall detection algorithm optimizes the threshold method.After testing on various public datasets and collected head pose datasets,the accuracy rate was improved from 93.9% to 99.64%.Finally,a physiological abnormality monitoring algorithm based on the headset system is designed and implemented.In order to solve the problem that physiological abnormal data is not easy to obtain,the algorithm extracts features from normal physiological signals through Auto Encoder(AE),and then uses One Class Support Vector Machine(OC-SVM)for abnormal detection.And add an attention-based two-stage long short-term memory network(Dual-Stage Attention-Based Long Short-Term Memory,DA-LSTM)in AE,by referring to the hidden state of the LSTM at the previous moment,in the encoding/decoding stage,respectively.Adaptively guides the weight assignment of the input sequence and the hidden state of the encoder in the selection time step,alleviating the problem of rapid performance degradation caused by AE when dealing with long time sequences.The experimental results show that in the case of using only normal physiological data,the algorithm proposed in this thesis can better detect abnormal physiological indicators,and its accuracy rate is 95.3%.