結合氣壓傳感器與加速度傳感器之跌倒偵測方法

Abstract

全球目前正面臨人口快速老化的問題。根據臺灣衛福部統計，2013至2015年65歲以上老年人口所有意外事故死亡事件中，跌倒為第二大主要死因；又根據美國疾病控制與預防中心統計，2012至2013年，65歲以上老年人口所有意外事故死亡事件中，55%為跌倒所導致，而在2000至2013年，65歲以上老年人口每十萬人中，因跌倒而死亡之比率將近成長兩倍。由此可見，跌倒成為高齡化社會之醫療照顧課題中急需解決之問題。 本研究提出同時使用氣壓傳感器與加速度傳感器之跌倒偵測方法，進入跌倒偵測方法前，使用氣壓傳感器之資訊，進行是否有一定程度之高度變化之判斷，若有，則進入跌倒偵測流程，高度變化作為Trigger Key以降低跌倒偵測所進行之較複雜運算。跌倒偵測流程分為兩階段，第一階段使用氣壓傳感器之資訊，並使用門檻值法進行是否進入第二階段之判斷，第二階段使用加速度傳感器之資訊，使用leave one-subject out進行訓練樣本與驗證樣本之分類，再使用支持向量機法進行是否為跌倒事件之判斷。 本實驗設計共有兩種情境：第一種情境為靜態之情況下執行動作之模擬，共24位受試者，執行日常生活動作與跌倒動作共5種；第二種情境為動態情況下執行動作之模擬，共9位受試者，執行日常生活動作與跌倒動作共15種。經過本研究所提出跌倒偵測方法之分類結果，第一種情境之敏感度、特異度與正確率分別為0.92、1、0.95；第二種情境之15種動作之敏感度、特異度與正確率分別為0.87、0.95、0.9，排除第15個跌倒動作，僅計算14種動作之敏感度、特異度與正確率分別為0.99、0.95、0.97。

Aging society is a global issue and growth of elderly population is growing faster. According to Ministry of Health and Welfare of Taiwan, fall events are the second cause of accident casualty of the elderly (people aging 65 years or older) during 2013 to 2015. In addition, statistics by Centers for Disease Control and Prevention of the U.S. also shows that 55% of accident casualty of the elderly caused by fall events during 2012 to 2013. Moreover, mortality of the elderly due to fall events grows nearly twice during 2000 to 2013. Therefore, fall events become an important issue among aging health care service. In this study, we proposed an integrating fall detection using barometer and accelerometer. Our algorithm contains three stages: Trigger Key: using barometer and simple threshold method to detect whether there is a significant negative height change, Stage One: using barometer and simple threshold method to determine whether there is a possible fall event, Stage Two: using accelerometer and support vector machine method to classify whether it is a fall event. For experiment design, we have two different scenarios. The first scenario is static and including two daily activities and three fall events. The second scenario is non-static and including seven daily activities and eight fall events. Sensitivity, specificity and accuracy of static scenario are 92%, 100%, 95% and 99%, 95%, 97% for non-static scenario.