Sleeping Posture Recognition And Get-up Intention Prediction Method Based On Intelligent Bed

China's aging population is becoming increasingly serious,many elderly patients cannot receive comprehensive and effective monitoring,and daily safety monitoring is one of the key issues in an aging society.Meanwhile,long-term bedridden patients are accompanied with many complications.Keeping the same sleeping position for a long time will increase the risk of bedsore.For the characteristics of different diseases,if the correct sleeping position is not adopted,it may not only affect the quality of sleep,but also cause secondary damage.The identification of sleep postures can be used to monitor the duration of a particular postures and predict possible diseases in combination with the physical condition of the bedridden patient,often changing the bed position can play a role in the treatment and prevention of the disease.In addition,elderly patients may also fall from bed because of the interference of physical factors.In order to prevent such accidents from happening,it is an effective way to judge the patient's behavior by monitoring the posture on bed.Therefore,sleeping posture recognition is a hot issue worthy of study.Based on the intelligent bed robot independently developed by our laboratory,A sleeping posture recognition method and a getting up intention prediction system are proposed in this thesis.Firstly,the nine postures we defined previously are recognized,including 6 normal postures(supine position,prone position,left lateral position,right lateral position,left frizzy position,right frizzy position)and 3 get-up postures(half up position,left edge position,right edge position).An extraction method of human body pressure features based on K-means clustering is proposed,This method can find the center of pressure distribution region of human body and take it as the pressure characteristic,experimental result shows that this method improves the recognition accuracy by 1.67%.In order to enhance the generalization ability of the posture recognition method,volunteers with different body sizes were selected to try 9postures according to their own habits and 1440 samples were collected as a data set.Then the accuracies of five different supervised learning algorithms on the same data set,multi-hiddenlayer fully connected neural network model achieves the best recognition accuracy of 88.33%,compared with the normal posture,the recognition accuracy of the get-up posture is lower.Through the analysis of the experimental results,it is found that predicting the intention to get up only by lying in bed posture is may cause false prediction.Therefore,a blended get-up intention prediction system is proposed.The process of human getting out of bed is a process that the pressure value continues to decrease.However,in reality,Frequent changes in physical exertion due to physical function problems and the strength of supporting the bed with both hands,or the action of turning over,these situations will decrease pressure and may cause the pressure value to fluctuate up and down,Therefore,in order to solve the problem that the pressure threshold is difficult to select,the pressure reduction process is identified based on the comparative analysis of the time series similarity,and finally combined with the sleeping position recognition results to form a blended get-up prediction system.When the pressure reduction process and get-up posture are recognized,the intelligent bed user has the intention to get up.Finally,a get-up behavior alarm system and a get-up experiment are designed to verify the effectiveness of the blended get-up intention prediction method to identify get-up intentions.The experimental results show that the method can accurately identify get-up behavior intentions.