An Indoor Video Surveillance with Intelligent Fall Detection Capability

[[abstract]]本文提出一個室內視訊監控系統，主要用來偵測人體的跌倒，此系統能偵測及判斷人體的姿勢狀態，並在人體跌倒時，發出求救訊息通知救護人員。為了能夠判斷人體移動的姿勢狀態，我們首先利用codebook建立背景模型的方式來建立背景資訊，並使用背景相減法配合陰影去除的方法，擷取可能移動物體的位置，由於擷取出來的前景影像可能包含雜訊或目標物內部出現破損，因此使用型態學濾波和尺寸濾波器來過濾雜訊並修補內部破損，當目標物被擷取出來時，我們以目標物的人體外觀比例以及人體高度等資訊來判斷姿勢狀態是否改變，當偵測到姿勢改變時，我們擷取目標物之投影柱狀圖來表示其形狀，並當作最近鄰居演算法K-Nearest Neighbor(K-NN)的輸入向量，用以判斷目標物的姿勢狀態，本系統經實際驗證，可有效的偵測人體不同的姿勢，並提高跌倒偵測率。由實驗結果得知，我們所提出之利用人體外框比例和人體高度位移比例來偵測姿勢狀態是否改變的方法能夠有效的提升整個系統的效能和辨識率。[[abstract]]In this thesis, we design an indoor video surveillance system for detection the falls of a human. The system can effectively detect and judge the posture of a human. When a person falls down, the system can detect and send an urgent message to inform the medical staffs. In order to judge the movement of a human, we adopt the method of codebook to build the background model, and utilize the method of background and shadow eliminations to extract the possible position of moving objects. The extraction of a foreground image will produce more noise and damage in this image. We also utilize morphological and size filters to eliminate the noise and repair the damage of this image. When the image object of a human is extracted, the aspect ratio and height of human body will be used to judge whether the posture is changed or not. In the meanwhile, the system detects a change of the posture and extracts the histogram of the object projection to represent the appearance. It will become the input vector of K-Nearest Neighbor (K-NN) algorithm and be used to judge the posture of the object. The method can effectively detect different postures of a human, and improve the accuracy of fall detection. In the thesis, we also propose a new method that uses the frame ratio and the displacement of height in the image to detect the posture. The experimental results show that our new method can further improve the performance of the system and the accuracy of posture identification