人體姿勢偵測系統與無監督辨識學習

A Sensing System of Human Postures and Its Unsupervised Learning of Pattern Recognitions

本文介紹一種穿戴式生理訊號偵測系統與相應的無監督學習分析技術。硬體設計方面採取三軸重力加速度感測器搭配神農無線藍芽傳輸模組。分析技術則採取量子力學中的密度泛函理論，使得數據中各特徵間的邊界以及數據群數能被同時決定。本技術針對長期實驗可用以偵測與分析人體姿體與各式時序生醫訊號的改變與關聯性。短期實驗可用以意外事故肇因分析以及警訊通知。除可節省大量人力資源與數據儲存空間的佔用，同時具備更高的可信度與客觀性，乃至於爾後的技術商業化。 The article proposes a wearable physiological signal sensing system associated with an unsupervised learningtechnique. The hardware mainly includes a three-axis accelerometer and a XenonBlue wireless transmissionmodule. Then the framework of density functional theory of the quantum mechanics was employed to extract thecluster boundaries of datasets and their corresponding cluster numbers. The proposed system provides requiringdetections and correlation analyses between the changing human postures and the time-dependent biomedicalsignals for the long-term experiments. The technique also can be used for real-time detections and analyses foraccident circumstances and then dangerous alarms. In a nutshell, the system not only provides highly plausibilityand objectivity but also reinforces the commercialization.