Research And Implementation Of Fall Detection And Alarm System Based On Android

The physical and mental health problems of the elderly caused by falls have aroused widespread concern.As falls may have bad implications for the elderly themselves,their families and the society,it is particularly important to deeply study and realize the fall detection and alarm program.At present,the study of wearable fall detection devices is the mainstream research of the fall detection in human activity recognition.However,the deficiencies in the wearable fall detection devices are its high cost and inconvenience.While as Android smart phones are widely used,many of their functions can be applied in the design,such as the built-in sensor,the capabilities of its strong computing,the stable and accurate wireless communication and location.Therefore,this thesis has formulated the practical design and implementation of a fall detection and alarm application based on Android.The specific study works are as follows.(1)First of all,in order to build data sets of normal human activities and falls,a data acquisition application has been developed,and data sample collecting programs have been designed.Then,the raw data samples are processed by adding window and median filtering.By analyzing the fall procedure,some eigenvalues which can distinguish falls from normal activities are extracted.With the eigenvalue normalization processing,the effective and regular feature data sets are collected,which provides solid data support for the fall detection algorithm design and simulation.(2)Next,a dual fall detection mechanism composing of the multi-parameter threshold method and SVM classification algorithm is designed.During the design of the multi-parameter threshold method,a lot of experiments have been conducted to find out the differences of falls and normal activities in acceleration and angular velocity.After that,synthetic vector acceleration SMA,synthetic vector angular velocity SMV and velocity V are extracted as threshold parameters in the design.The multi-parameter threshold method with SMA,SMV and V are used as the primaryalgorithm to quickly detect and filter normal activities which are obviously different from falls and improve the efficiency of falling-action detection.While in the design of SVM classification algorithm,LIBSVM is used to train the data sets.Model optimization strategy is adopted to construct the well-performed classification models which are used to correctly distinguish falls from the suspicious fall events detected by the multi-parameter threshold method.Finally,the detection effect of the dual fall detection is verified by MATLAB.The experimental results show that the fall detection algorithm in this thesis has high sensitivity and specificity.(3)The above dual fall detection mechanism is applied in the fall detection and alarm system on Android.The main functions of the system are designed and realized,including data management,fall detection,positioning,alarm and so on.The system has been tested and analyzed according to the established experimental program.The results show that the system is with high detection accuracy and real-time performance.