對網際網路的各種應用而言,提供一個逼真、具吸引力及含有豐富內容的網頁是必要的。因個人電腦運算速度加快、網路頻寬加大及網路3D語言快速發展,建構一個互動、栩栩如生的3D虛擬環境成為未來趨勢。由於真實環境中都會有其它的人環繞在四周,對建構一個真實、動人的3D虛擬環境而言,加入虛擬人物是必要的。由於真實人物具有許多的關節點,且每個關節點又具有多個自由度,故決定人物的姿勢是非常困難。雖然動作擷取方式可以有效決定人物的動作,但目前的設備昂貴、安裝麻煩及所搭配軟體學習不易等因素,故發展一套低成本、容易操作的動作擷取系統,以促進發展一個普及化的虛擬環境是迫切需要的。根據圓錐曲線理論,一個已知半徑3D平面圓形的位移及旋轉量可經由它的2D橢圓投影推導出。利用此特性,本研究在表演者重要的追蹤點貼上圓形標記,使用單台攝影機擷取表演者的移動影像,使用界限框快速計算出在影像中的橢圓。利用平滑假設的特性,可以分析界限框的方式估算部份覆蓋的橢圓,接著計算圓形標記的3D位置,以決定虛擬人物的姿勢。換言之,表演者在攝影機前擺設動作,以”操作玩偶”的方式控制在虛擬環境中虛擬人物的動作。
A vivid, attention-holding and content-rich homepage is an essential part for many internet-based applications. With the rapid advance in software support, i.e., the internet 3D scene description language VRML and Java 3D, and in hardware support, i.e., more powerful PC and larger bandwidth, the current text and 2d image centric homepages are soon expected to include interactive, animated 3d virtual environment. For a 3d virtual environment to be realistic and appealing, it must contain humans, as in real world we rarely enter a place that does not have people. Controlling the movement of a virtual human, however, is overwhelmingly complicated due to the large number of human body joints and the many numbers of degree of freedom associated with each joint. Although the motion capturing techniques do provide an efficient way to animate complicated human actions, they require cumbersome and expensive devices that are difficult to set up and learn. To facilitate the development of a human-popularized virtual environment, a low-cost, easy-to-operate motion capture method is developed. From conics theory, the 3d rotation and translation of a planar circle with known radius could be derived from its 2d elliptic projection. Based on this property, we attach circular marks to the important tracking points of a human user and capture the movement of the human performer using a single video camera. Elliptic projections of these marks in the captured image sequences are quickly traced using their bounding boxes. Based on the smooth-movement assumption, the partially overlapped ellipses could also be determined by analyzing their bounding boxes. After calculating the 3D positions of the circular marks that represent the joints of the virtual human, the composing parts of the human model could be rotated properly to let the virtual human follow the human player’s movement. In other words, the user performs in front of a video camera to “puppeteer” the virtual human in a virtual environment.