Research On Sleeping Posture Recognition Algorithm Based On Multi-Channel Body Vibration Signal

With the development of society and the abundance of living conditions,people pay more and more attention to their sleep quality.Sleeping position monitoring plays an important role in sleep health care and intervention.In terms of medical diagnosis,sleeping position is an important indicator for the diagnosis and treatment of postural sleep,respiratory,heart and cardiovascular diseases.Accurate identification of sleeping position is of great significance for the diagnosis of human sleep disorders.At present,the traditional sleeping position monitoring equipment is only used in the field of clinical medicine,requiring professional medical staff to operate and maintain,with high cost,complex operation and other shortcomings,so it has not been popularized for civilian use.At present,with the acceleration of the aging society,the potential sleep diseases of the elderly have also increased the medical burden of the family,and the monitoring of sleeping position can prevent and control some chronic diseases in advance.Therefore,a set of light and simple sleeping position monitoring system has a broad market.Based on the traditional sleeping position recognition,this thesis proposes a sleeping position recognition method based on body vibration signals.Body vibration signals contain abundant information,such as cardiac impact signal(BCG),respiratory signal and so on,which are also closely related to sleeping position.Firstly,non-contact intelligent mattress was used to collect the original body vibration signal of human body,and the largest part of the body vibration signal was analyzed through the spectrum diagram.The strength of the body vibration was quantified,and the distribution of the body vibration was extracted as the characteristics of the breathing mode.Meanwhile,the time domain features of BCG signal are extracted as auxiliary features of BCG mode.Sleeping position recognition is a multi-classification recognition problem in essence.In this thesis,multi-modal fusion is carried out based on two modes of breathing and BCG.Based on the characteristics of breathing signals,multi-classification Logistic algorithm,namely Softmax algorithm,was used to calculate the mapping relationship between breathing characteristics and sleeping position.Based on the characteristics of BCG signal,the mapping relationship between time domain characteristics of BCG signal and sleeping position was calculated by random forest algorithm.The combination of the two models with stacking improved the algorithm’s recognition rate for both left and right sleeping positions.At the same time,we also did a sleeping position recognition experiment based on human body position pressure.A thin film resistance sensor was used to collect body position data,and the distribution matrix of body position pressure was constructed.The mapping relationship between body position pressure characteristics and sleeping position was obtained by machine learning model.Compared with the fusion model,it is found that the body motion changes have a greater impact on the pressure distribution of body position,but a smaller impact on the strength distribution of body vibration.The sleeping position recognition algorithm based on body vibration signals can effectively suppress the impact of body motion on the prediction results.Compared with THE SVM support vector machine model and the mixed Gaussian model,the model proposed in this thesis has a certain improvement in recognition accuracy on the basis of real-time performance.