隨著科技的進步，監視系統的安裝也愈來愈廣泛，要檢查擷取下來的畫面中是否發生特定事件，通常需要很長的時間及大量的人力介入。為了解決這個問題，目前已有許多可自動判斷視訊物件行為的方法被提出來，比起其他方法，以移動歷史影像為基礎的方法通常具有計算複雜度低與容易實作等優點，因此也比較受大家的歡迎。本篇論文提出一個快速移動歷史影像方法，針對每一已知行為事先建立多組特徵值、運用部分距離計算法、以及變換區塊計算順序，來降低移動歷史影像方法的運算時間。為了驗證所提方法的有效性，本篇論文採用Chen方法中九個區域的平均像素移動方向作為特徵值，並以歐基里德距離作為相似度比對。依據實驗結果顯示，所提方法確實能夠有效降低運算時間。 As the progress of technology, the surveillance system is installed more and more widely. It usually requires a lot of time and human efforts to check if a specified event occurs in the captured video. To solve this problem, there are many approaches were proposed to recognize behaviors of a video object automatically. Among available behavior recognition methods, the MHI-based approaches are more popular for they have less computational complexity and are easier than another. In this thesis, a fast MHI approach is proposed to reduce the computation time of the MHI approach by storing the multiple sets of features for a predefined behavior, using the partial distance computation method, and changing the calculated order. Nine local information proposed by Chen and squared Euclidean distance are used in the behavior matching process in this thesis to manifest the performance of the proposed approach. Experiment results show that the proposed method can effectively reduce the computation time of MHI approach.
