從連續影像中擷取人體的運動

Abstract

在此篇論文中﹐我們提出了一個新的演算法﹐從一連串的影像中﹐擷取人 體的運動。我們的演算法是以"點"及"特徵"來做為擷取的技巧。這個演算 法﹐主要由三個步驟組成。首先﹐由連續的影像中﹐找出人體移動的資訊 。其次﹐利用此資訊﹐來發展出一個完整的人體運動。最後再對此人體做 一更精確之修正。我們提出的演算法﹐具有簡單及容易實作的特性。從實 驗結果來看﹐我們的演算法﹐的確可以清除掉一張影像中﹐大部份的靜態 背景﹐剩下運動中的人體被保留。 In this thesis, we present an algorithm to extract human motion from a sequence of images. The proposed algorithm can be considered as a combination of pixels based and features based segmentation. The algorithm consists of three steps: (1) extraction motion information. (2) region growing. (3) refinement. Our algorithm is easy and practicable. The scheme has been applied to a number of realistic sequences with encouraging results. Although the technique is developed for human body motion, it is speculated that the same algorithm can be applied for segmenting other types of sequences with moving objects.