Research On Fall Detection Algorithm In Elderly Health Management

The increasing degree of aging has made the health of the elderly one of the social issues.In the health management of the elderly,falls have become the main influencing factor of accidental injuries or casualties among the elderly.Failure to provide timely assistance when a fall occurs will lead to irreparable harm.Therefore,how to automatically detect falls using advanced assistive technology has become a hot research area.Wearable devices have become a popular carrier for users to monitor their physical behavior in real time due to their low cost and easy portability.Therefore,it is of great theoretical and practical value to study the automatic fall detection algorithm in a wearable environments.In the past,the threshold detection method commonly used in fall detection algorithms will be affected by different groups of people,and cannot be applied to the same threshold interval,resulting in poor recognition results.Machine learning algorithms rely on artificial feature extraction and effective feature selection,which makes classification difficult.Compared with other traditional machine learning models,deep learning has the advantage that it does not require empirical knowledge,but can autonomously learn local and global features of data signals,which has attracted much attention in the field of fall detection.Therefore,this paper deals with the global dependence of long-term sequence data on convolutional neural network,long short-term memory network models that are complicated to calculate and time-consuming to train,and are difficult to apply to the defects of mobile wearable devices.FD-GRU deep network model with global characteristics and capable of shortening the delay.At the same time,make full use of and enhance the time dependence of processing time series data and take into account the influence of data space to ensure that the model has low complexity,and improve its ability to identify multiple types of activities and activities with high signal similarity.The triaxial acceleration sensor data in the Mobi Act Dataset was selected to verify the validity of the model.The accuracy,recall,specificity,precision,and F1 values were 99.51%,96.47%,99.74%,96.22%,and 96.28%,respectively.In the forward direction.,lateral,and backward falls all show good recognition results,and can accurately identify the other 9 daily behavior activities,the indicators are not only better than the traditional machine learning models such as Bayesian algorithm,J48 decision tree,K nearest neighbor algorithm,random forest and support vector machines,and better than deep learning models such as CNN and LSTM.This technology can be used to monitor the activity of the elderly in real time on the mobile terminal and assist the alarm to notify the family and medical staff to prevent secondary injuries,providing new ideas for the elderly’s health management and smart medical treatment.