本研究是基於在考量被照護者隱私狀況，以較低成本的低解析度熱感應器，配合深度學習的方式，進而監控被照護者之床旁狀況，包括跌倒偵測、離床警示、翻身狀況及睡眠品質。本設計的研究中使用的低解析度熱感應器安裝於病房上方，輸出圖像raw data 後再轉成圖像，再搭配實際攝像機做標識(Labeling) 判斷，最後傳至類神經網路(CNN) 做模型學習。本論文使用自行架設之環境實錄進行模型訓練，經過整理後取得10144 張圖像。此資料集80% 用來訓練模型，20% 用來驗證模型，模型調整後可達87.92%以上的準確率(accuracy radio)。;This study aims to monitor the bedside conditions of caregivers, including fall detection, bed exit alerts, turning over, and sleep quality, using low-resolution thermal sensors that are more cost-effective and consider caregiver privacy. The low-resolution thermal sensors used in this design are installed above the patient room, output image raw data, and then converted into images. They are then labeled and judged using an actual camera and finally transmitted to a deep neural network (DNN) for model training. This paper uses the data from the self-built environment for model training and obtains 10,144 images after整理. 80% of this dataset is used to train the model, and 20% is used to verify the model. After the model is adjusted, the accuracy ratio can reach over 87.92%..
