基於影像處理及卷積式類神經網路之居家照護系統

Abstract

近十年人口老化的問題越來越突顯。以台灣為例，出生人口逐年的下降，已成為世界人口成長的末段班。有些研究及報告預估不久的未來，台灣會因為經濟狀況以及其他因素的影響，死亡人口成長率將會超越出生人口的成長率，更加速了台灣人口老化的速率。未來高齡化的腳步越來越近，相關的政策制定及科技的輔助變得十分重要。 隨著數位科技越來越發達，電腦技術的進步。不管在半導體製程上或者是軟體工程及演算法的改進上。運算速度比起以前得到跳躍性的提升。從單核心運算元，一路到多核心運算元。圖形處理器的效能提升及相關輔助軟體的開發。造就以往過去十幾年因為需要大量運算而難以運行的神經網路又重新受到了世人的重視，隨著卷積式神經網路的發展，影像處理的技術也得到顯著的成長。 本論文提出一個基於影像特徵，並利用影像處理及類神經網路之居家型照護系統。希望利用影像處理能達到監測自動化，藉此提升老年人居家的安全性。

Population aging has become an important issue in the past ten years. Taiwan as an example, the birth rate decays year by year and has become the bottom of the world. Depending on Taiwan’s economy and other factors, some researches had shown that our death rate will exceed birth rate and further accelerate population aging problem. To face population aging problem, relative policies and assistive technologies become very important. As the great progress in both semiconductor manufacturing and algorithm, computational speed had grown rapidly. Artificial neural network(ANN) was a tough issue due to the lack of computation power. But as the advance of our hardware, from unit processor to multi-processors, GPU performance improvement and relative auxiliary software. These had led artificial neural network(ANN) regained attention. Later as convolution neural network(CNN) flourished, image processing technology also grown significantly. In our thesis, we proposed a home care surveillance system based on image processing and artificial neural network. We expect to achieve automatic monitoring to enhance elder home care safety.