「健康保全」 U 化遠距健康監測與自我關懷照護系統---以視訊為基礎的日常生活與與配合醫療行為監測之U化健康照護系統

Abstract

本計畫發展一個以視訊為基礎的日常生活與與配合醫療行為監測之U 化健康照護系 統。 在居家、老人安養、復健中心或病房、慢性病、健康照護站等，每一個人日常生活 行為與活動，包括 起床、上廁所、洗澡、吃飯、休閒閱讀與書寫、使用電腦等活動；配 合醫療行為，包括 投(服)藥、血壓、体溫、心電圖、血糖等生理訊號測量，均是健康照 護的核心。 藉由數隻攝影機，經前背景分離與動作辨識等視訊處理技術，可建立上述行 為與動作判斷與識別，獲得其日常生活活動正常曲線輪廓 (behavior profile)。當一個人正 常生活行為有偏離其正常曲線輪廓、有偏差醫療行為、或有異常生理訊號數值時，一方 面可以定時提醒病人注意，再進行生理訊號的量測，一方面並將監控與量測的結果與確 實用藥的時間等資訊利用無線通信裝置回傳至伺服器作後續的判讀與處理，經伺服器判 讀後上傳至醫護端。 本計畫發展一個以視訊為基礎的之U 化健康照護監測系統，前、背景分離 與動作辨 識 等視訊處理技術為本計畫的智慧中樞，是為第一年計畫研發的重點。第二年專注於日 常生活行為與活動正常曲線輪廓的技術建立，並對健康危害很大的 跌倒，包括昏倒，預 防與偵測的技術研發。 第三年計畫研發的重點項目，包括 投(服)藥、生理訊號確實與規 律測量 之配合醫療行為的動作判斷，及全系統整合與測試。

A fully automated, passively activated vision-based system will be developed to allow routine, continuous, non-obtrusive monitoring of selected Activities of Daily Living (ADL) and the production of a behavioral record that could be subjected to trend analysis. In this subproject, we will examine at-home activity rhythms and present behavioral patterns obtained from an activity monitoring study in an assisted living setting. Established behavioral patterns have been captured using custom software based on a statistical predictive algorithm that models circadian activity rhythms (CARs) and their deviations. The CAR was statistically estimated based on the average amount of time a resident spent in the rooms including bedroom, bathroom, kitchen, living room etc., within their assisted living apartment, and also on the activity level given by the average number of motion events per room. A validated in-home monitoring system (IMS) recorded the monitored resident’s movement data and established the occupancy period and activity level for each room. Together with the sensing of various physiological signals as well, such as skin conductance, blood pressure, respiration rate, and electrocardiogram (EKG) measurement, they can potentially aid in assessing the health status of the monitored resident. Using these data, residents’ circadian behaviors and physiological signals were extracted, deviations indicating anomalies were detected, and the latter were correlated to activity reports and notes generated to facility’s professional caregivers on the monitored residents. The system could be used to detect deviations in activity patterns and to warn caregivers of such deviations, which could reflect changes in health status, thus providing caregivers with the opportunity to apply standard of care diagnostics and to intervene in a timely manner. The aim of this subproject is to develop an in-home video ubiquitous health monitoring system which is capable of measuring individualized health status through one’s ADL profile and physiological signals reporting the abnormality to the primary care provider and caregivers alike, thereby allowing timelier and targeted preventive interventions. The capturing of the resident’s posture binary image based on the generated background model and activity recognition will be the main research topics of the first year of the project. The second year of this project research will be directed to the developing the tool to develop behavior profiles from the resident ADL. A fall detection technique and a fall incidence prediction system will also be investigated. Drug dose compliance monitoring unit and physiological signal measurement validation component will be studied in the third year. Moreover, pilot testing and evaluation to integrated our system to other subproject modules will also be conducted.