Research And Implementation Of Early Warning Method For Elderly Falling Risk Based On Multi-dimensional Data Fusion

With the acceleration of the aging process in China,the health and safety issues of the elderly have gradually attracted people’s attention.Fall is one of the most dangerous accidents that threaten the health and safety of the elderly.In response to the fall threat of the elderly,the existing research has mainly focused on fall detection,which is used for emergency treatment after a fall,and belongs to a post emergency strategy.Once the damage to the elderly,families and society caused by falls is done,and the remedy measures are extremely limited.Therefore,how to predict the risk of falls through the estimate of physical state of the elderly,formulate the corresponding active intervention strategies,reduce the probability of falls in the elderly,and fundamentally protect the health and safety of the elderly are the most effective ways to deal with falls of the elderly.At present,the research of falls risk assessment for the elderly focus on gait analysis and posture detection,and a large number of risk assessment models have been obtained.However,the construction of most risk assessment models lacks daily life data under real scenarios,which cannot assess the physical status of the elderly effectively,and the model is lack of effectiveness.In addition,most studies use a single device to perceive the physical state of the elderly.Noise and occlusion problems are insurmountable difficulties for a single perception assessment model.At the same time,the intrusiveness of the sensing device brings about the problems of subconscious resistance and data credibility.In view of the problems above,this paper studies the low-invasion perception analysis algorithm for the behavior of the elderly from the perspective of multi-dimensional perception redundancy complementary,to achieve a high-precision assessment of the risk of falling.The specific research contents are as follows:First,a walking stability analysis model based on gait analysis is constructed.Walking is one of the most basic functions of the human body,and gait stability analysis is one of the most effective methods of fall risk assessment.Aiming at the highly invasive problem in gait analysis,in this paper,the advantages of high accuracy and low intrusion of lidar are used to perceive the daily gait of the elderly,to study the precise tracking and the recovery of occlusion,and to realize the accurate analysis of the stability of the daily gait of the elderly.Secondly,this paper constructs a walking balance detection model based on posture analysis.Body balance has a high correlation with falls,and the detection of walking stability is of great significance to the assessment of fall risk.Aiming at the problem of privacy invasion of traditional RGB camera and the loss of bone posture caused by occlusion in posture detection,in this paper,we use the depth lens which does not violate the privacy to capture the elderly’s walking posture,and study the elderly’s posture detection and occlusion recovery based on adaptive angle of view rotation to achieve the detection and analysis of the daily walking balance of the elderly.Then,the detection model of swing arm balance based on walking selfcorrelation analysis is constructed.Arm swing is a regular behavior in the process of walking.Balance detection of swing arm is one of the effective methods to evaluate the risk of falls.Aiming at the accuracy loss caused by occlusion in the analysis above,in this paper,with the features of low intrusion and continuous monitoring of smart watch,the balance of the swing arm of the elderly is analyzed around the clock,and the information loss and prediction error caused by occlusion are supplemented and corrected.After that,this paper constructs a fall risk assessment model for the elderly based on multi-dimensional data fusion.Aiming at the problem that single data fall risk assessment is vulnerable to environmental impact and has low accuracy,from the perspective of redundancy and complementarity,multi-dimensional fusion of gait features,posture features and swing arm balance features is carried out,and an attention mechanism model combining Gated Recurrent Deep Neural Network(DNN)is proposed.The experimental results show that it can evaluate the fall risk of the elderly in daily walking correctly,with an average accuracy of 84.7%,which proves the effectiveness and accuracy of redundant and complementary data fusion in the fall risk assessment.Finally,an early warning system of fall risk for the elderly based on multidimensional data fusion is designed and implemented.The data analysis mentioned above and fall risk assessment are integrated and tested and the results verify the feasibility of the above-mentioned research methods and theories.