Research On Fall Detection Method For The Elderly Based On Accelerometer From A Smart Watch

With the aging process of the population increasing,the number of the elderly living alone and empty nest is increasing.Elderly accidental falls at home or when they go out have become the main problems that cause the accidental injuries and even directly affects the health and life safety of the elderly.Therefore,it is of great social significance and application value to develop a monitoring system with high accuracy,real-time monitoring of elderly falls and early warning function,and easy to carry.In recent years,wearable devices are widely used in falls detection of the elderly for their advantages such as convenient carrying,real-time health monitoring anytime and anywhere,sensitive sensors,long standby time and low price.Among wearable devices,the wrist wearable device for fall detection is more easily accepted by the elderly due to its comfort.However,due to the interference of frequent random movements of hands and wrists,the accuracy rate of fall detection based on wrist wearable device is difficult to reach95%.Aiming at the problem of low accuracy of fall detection in wrist wearable devices,this paper takes wrist-worn smart watch based on triaxial accelerometer as the research object.Firstly,the daily activities of the elderly are classified.On the basis of considering the influence of hand and wrist activities,the experimental scheme is carried out.Young and middle-aged volunteers are selected for the experiment to obtain the experimental data.Then,the threshold fall detection algorithm and the intelligent fall detection method ESAEs-OCCCH based on accelerometer data from a wrist-worn smart watch are proposed.Finally,the feasibility and superiority of the two methods are verified by comparisonThe main research work of this paper is organized as follows:(1)The behavior of the elderly is classified,the experimental scheme is designed,and the experiment and signal are acquired.Human activities are divided into two categories: ADLs and FAs.Based on the cause of the fall,the behavioral characteristics of the elderly,the uncertainty of ADLs and FAs,and the influence of the intense activities of the hand on the accelerometer Signal,thirteen FAs and sixteen ADLs are designed in the experiments.Young and middle-aged volunteers of different ages,heights and weights are selected to carry out experiments on each type of activity and obtain accelerometer signals.Analysis of the collected data samples,it shows that the fall accelerometer signal has impact and concussion,and the impact concussion intensity is related to the strength of the volunteers hitting the ground when they fall.Among the sixteen ADLs,the most obvious impact phenomena are knocking on the table?applause?quickly sitting down?and quickly standing up,followed by running,waving and exercise.(2)Based on the impact,positive and negative and concussion characteristics of fall acceleration signal,a threshold fall detection method is proposed.Firstly,the characteristic parameter of the sum of the triaxial acceleration range is used to reflect the impact of the fall signal.Then,the triaxial sign function and the sum of zero crossing points are designed and put forward to reflect the positive and negative,and concussion of the fall signal.Finally,the kurtosis of the resultant acceleration envelope spectrum is designed and put forward to distinguish the fall behavior and the knock on the table behavior.The thresholds of the four characteristic parameters are obtained through statistical analysis and many attempts.The experimental results show that the sensitivity of the proposed algorithm is 99.991%,the specificity is 92.31%,the gmean is 96.10%,and the average time to detect a sample is 0.0168 seconds,which verifies the accuracy and efficiency of the proposed method.(3)Combining ESAEs with OCCCH,an intelligent fall detection method named ESAEsOCCCH is proposed.Firstly,ESAEs is adopted for unsupervised feature extraction.Then,OCCCH is used for pattern recognition.Finally,the majority voting strategy and weight adaptive adjustment strategy are combined to improve the performance and stability of the algorithm.The ESAEs-OCCCH method is compared with twelve SAEs-OCCCHs?OCCCH with statistical features statistical?OCSVM with statistical features statistical,SVM with statistical features statistical and KNN with statistical features statistical.The experimental results verify the accuracy,feasibility and stability of the ESAEs-OCCCH method.(4)The proposed the threshold fall detection algorithm and ESAEs-OCCCH method are compared and analyzed in terms of accuracy,feasibility and reliability.The experimental results show that the threshold fall detection algorithm has higher sensitivity,but the specificity,gmean and algorithm reliability are lower than ESAEs-OCCCH method.In addition,although the efficiency of the two methods can meet the practical needs in the actual situation,the testing time of ESAEs-OCCCH method is faster than that of threshold fall detection algorithm.Therefore,generally speaking,the ESAEs-OCCCH method are better than threshold fall detection algorithm.The research work of this paper has certain theoretical significance and engineering application value for the detection of falls of the elderly,the weak and the disabled.