Feasibility of real-time detecting and preventing related injuries of unexpected falls using smart mobile devices
|關鍵字:||智慧行動裝置;非預期跌倒;即時辨識;smart mobile phone;unexpected fall;real-time recognition|
儀器使用具備三軸加速規與陀螺儀晶片的智慧型手機作為感測器，招募12位健康男性受測者(1.69±0.03 m; 67.33±7.52kg; 24.30±2.02 yrs)於模擬居家實驗空間進行日常生活動作實驗，分別為坐、坐到站、蹲、蹲到站、上下床、走、上下樓梯與慢跑各10次試驗，並建置絆倒與滑倒步道，進行正常速度行走中各1次非預期絆倒與滑倒。搭配手機作業系統開發程式紀錄各動作三軸加速度與角速度後分析疊加數值與跌倒時間。跌倒閾值設定為三軸疊加加速度4.6m/s2與人體向前與側向翻轉方向角速度疊加3.6rad/s，辨識跌倒的靈敏度為90.24%，特異性為94.75%，用於觸發防護系統之領先時間絆倒為173.00±52.80毫秒，最小值為101毫秒，排除坐下型滑倒後為156.50±18.34毫秒，最小值為135毫秒，95%信賴區間為(119.82,193.18)，而坐下型滑倒僅為94.71±27.96毫秒，最小值為51毫秒，95%信賴區間為(38.79,150.63)，範圍符合市售防護氣囊最小觸發時間35毫秒範圍。設定100毫秒做為觸發防護時間防護絆倒與躺下型滑倒，坐下型滑倒至少50毫秒之預先防護效果。招募30位男性受測者(1.69±0.04 m; 63.17±7.37kg; 24.36±2.04yrs)測試即時跌倒辨識，進行非預期絆倒與滑倒各一次，辨識訊號由聲響表示，絆倒靈敏度為100%，滑倒為91.67%。時間分析中，絆倒與躺下行滑倒領先時間為175.96±53.07毫秒，最小值為100毫秒，95%信賴區間為(69.81,282.10)。坐下型滑倒領先時間為107.71±45.51毫秒，最小值為67毫秒，95%信賴區間為(13.68,195.74)，部分坐下滑倒可能小於市售35毫秒防護，故皆須採用預先防護效果。結果顯示跌倒辨識靈敏度與特異性皆高於90%，擁有良好使用成效，且本研究達成使用智慧行動裝置即時偵測非預期跌倒，並提供過去未針對的非預期滑倒辨識進行分析。即時辨識實驗中驗證一般使用者使用時亦效果良好，保護時間大多亦能符合市售防護氣囊最小觸發時間35毫秒範圍，確立可用於即時防跌。|
Injuries and deaths always occur in the elderly when falls happened. The utilization of inertial measurement units (IMU) for fall detector could recognize fall and even prevent injuries caused by fall impact during activities of daily living (ADL). Studies have tried methods of IMUs using, real-time detecting in falls and triggering unexpected falls. They were researched separately so the present study tries to use a smart phone which includes IMUs and operating system to act as a detector with fall recognition, trigger signal output and real-time recognition. This is because it is highly portable and easy to use. This research also focus on real-time slip detection those previous studies did not mentioned. 12 young healthy subjects (height: 1.69±0.03m; weight: 67.33±7.52kg; age: 24.30±2.02yrs) performed ADL and encountered unexpected falls. Trunk motion through acceleration (in accelerometer) and angular velocities (in gyroscope) were measured through a smart phone. The acceleration and velocity of trunk motion were measured by the accelerometer and gyroscope in smart phone. In fall experiments, subjects were asked to walk on a path with designed mechanisms to trigger trip and oily spilled in trigger slip. Thresholds in smart phone were set to 4.6 m/s2 in 3 axis superposition accelerations and 3.6 rad/s in pitch, roll superposition angular velocities. Time delay between 2 thresholds was set 225ms to ensure trip and slip are true. Falls distinguishing from ADL showed 90.24% sensitivity and 94.75% specificity. Lead time of trip and slip (lying down) is 156.50±18.34 ms. The minimum is135 ms. 95%CI is(119.82,193.18) ms. Lead time of slip(sitting) is 94.71±27.96 ms. The minimum is 51ms. 95% CI is (38.79, 150.63) ms. All of above lead time are smaller than 35 ms that can suit protector triggering on the market. 30 young healthy subjects (height: 1.69±0.04m; weight: 63.17±7.37kg; age: 24.36±2.04yrs) performed real-time fall recognition tests and the device showed 100% sensitivity in trips and 91.67% sensitivity in slips. The signal of recognition is presented as a voice. Lead time of trip and slip (lying down) is 175.96±53.07 ms. The minimum is100 ms. 95%CI is (69.81, 282.10) ms. Lead time of slip (sitting) is 107.71±45.51 ms. The minimum is 67ms. 95% CI is (13.68,195.74) ms. Pre-protection is needed because part of lead time smaller than 35 ms. The real-time recognition is corresponded to the experiment of the previous stage on ADL and falls distinguishing. Results show high sensitivity, specificity and enough lead time to trigger protector in unexpected fall through a smart phone. It not only showed high accuracy but provides convenience for general user. The system is better than previous studies which use less conditions and is suitable for those who living in a home environment. This research also analyzed unexpected slip to understand kinds of slip and their impact.