整合慣性感測器與肌電訊號量測之穿戴式跌倒感測器

Abstract

跌倒是高齡者事故傷害與死亡的主要原因，若能在跌倒發生時，快速且準確的偵測跌倒，有助於降低跌倒傷害，減少相關之醫療照護成本。慣性感測器已廣泛用於人體動作辨識，本實驗室先前研究則指出，肌電訊號有助於提升跌倒辨識速度。本研究承接過往之技術，藉由整合慣性感測器與肌電訊號量測，改善使用上之便利性，開發可穿戴式的即時跌倒辨識系統。 本研究之跌倒感測器，由設置於使用者腰部前方的三軸加速規與三軸陀螺儀，以及雙腳骨直肌上的機電訊號量測裝置組成。實驗共募集六位受測者(23.3±1.0 yrs; 168.7±6.4 cm; 60.70±3.98kg)，在模擬環境中執行日常生活與模擬跌倒動作。日常動作包含蹲、坐、走與撿拾物品等；跌倒則模擬高齡者最常見的絆倒，在受測者行進時，使用繩索裝置隨機干擾受測者左腳或右腳的平衡，使受測者跌落防護軟墊上。本研究使用肌電訊號與運動學資料，設立最佳化門檻值辨識跌倒。使用肌電訊號進行辨識時，由於量測參數有複數個，因此建立四種不同的整合辨識方法，並建立完整的統計評估流程，判斷何種辨識方式具有最佳效能。 使用慣性感測器辨識跌倒之Sensitivity為89.6%，Specificity為98.3%；肌電訊號之Sensitivity為72.9%，Specificity為92.0%，慣性系統辨識準確率較高，肌電訊號則在辨識速度具有優勢。整合兩者進行綜合判斷，Sensitivity為81.2%，Specificity為98.6%，雖然準確率比單獨使用慣性感測器時下降，但在辨識速度上卻有所提昇，領先時間為264±178 毫秒，增加約50毫秒。在95%信心水準下，Sensitivity之信賴區間下限仍超過70%，證實利用本研究開發之跌倒辨識方法，此穿戴式感測器已經可以準確的偵測跌倒。

Falls are leading causes of unintentional injuries and deaths in the elderly. To detect falls early and accurately is important for reducing fall-related socioeconomic cost. Inertial sensors have been used to distinguish fall from activities of daily living (ADLs). Our previous study found that using electromyography (EMG) signals on fall detection has advantage on recognition speed. Therefore, the aim of this study was to further develop a wearable fall detector by using inertial and electromyography combined sensors to detect fall events before impact. We have established a prototype of a fall detector, combing a tri-axis accelerometer and a tri-axis gyroscope, attached to the frontal surface of waist, and two-channel EMG sensors, recording the activities of bilateral rectus femoris muscles. Six subjects (23.3±1.0 yrs; 168.7±6.4 cm; 60.7±3.98kg) volunteered to the experiment for evaluating the performance of the fall detector. Each subject performed several ADLs in a mimic living environment and a few unexpected simulated trips, induced by a custom-made device attached to the ankle, were interspersed among the ADLs. Self-developed detecting algorithms were used to distinguish falls (trips) from ADLs, and a optimal detecting algorithm was then determined based on the detecting performance using a set of statistical analyses. Using inertial sensors alone could detect fall with 89.6% sensitivity and 98.3% specificity. By using EMG alone, sensitivity was 72.9% and specificity 92.0%. The fall detector using the combined sensor could identify falls before the impact between the human body and floor, with sensitivity of 81.2% and specificity of 98.6% with 264±178 ms mean lead time. To detect fall using inertial sensors are more accurate than using EMG signals alone. However the addition of EMG sensors to inertia sensors allowed our system to have about 50 ms more lead time. Besides, the lower limit of the 95% confidence interval of the system sensitivity was 73.0%, which may still be suitable for some real-life application. In conclusion, we have demonstrated the possibility of using EMG sensors combined with accelerometers to provide accurate and fast trip-fall detection.