|
Abstract
|
OBJECTIVE To establish an identification model for elderly people who are prone to falls, and to provide analytical methods for preventing and controlling elderly people from falling.
METHODS In this paper, we analyzed the kinematic data of the elderly collected by the timed up and go (TUG) test, and used the displacement of the centroid of each body segment (head and neck, arms, lower torso, legs) during the travel of the elderly Establish a neural network model for the feature dimension; collect the occurrence of falls of the test staff in the past year through a questionnaire, and divide the population into fall-prone and fall-prone according to the history of falls.
RESULTS The recognition model based on general regression neural network (GRNN) had good effect and the accuracy was 95.83%. Probabilistic neural network (PNN) and error backpropagation neural network (BPNN) obtained 87.50% and 75.00% recognition accuracy, respectively.
CONCLUSION The identification method of the elderly who fall easily based on neural network has a high accuracy rate, and the early recognition of the elderly who fall easily helps to reduce the occurrence of falls.
Keywords : fall , centroid displacement , neural network , recognition
摘要:目的 建立易跌倒老人的识别模型，为预防和控制老年人跌倒提供分析方法。方法 本文通过对起立行走计时（timed up and go，TUG）试验收集的老年人运动学数据进行分析，以老年人行进期间的各体段（头颈部、手臂、下躯干、腿部）质心位移为特征维度，建立神经网络模型；通过问卷收集试验人员在过去一年的跌倒发生情况，按照有无跌倒史将人群分为易跌倒和不易跌倒。结果 基于广义回归神经网络（general regression neural network，GRNN）为基础的识别模型效果好，准确度为95.83%。概率神经网络（probabilistic neural network，PNN）和误差逆传播神经网络 （backpropagation neural network，BPNN）分别得到了87.50%和75.00%的识别准确率。结论 基于神经网络的易跌倒老人识别方法得到了较高的准确率，易跌倒老人的早期识别有利于减少跌倒事件的发生。
关键词 ： 跌倒, 质心位移, 神经网络, 识别
Language: zh |