虛擬實境中以真人帶領木偶做肢體動作之電腦視覺技術

Abstract

本論文首先要解決目前虛擬實境（virtual reality）系統安置位置 追縱器 （ position tracker）在使用者的頭部或四肢上，需靠電線傳送 追蹤數據到電 腦，造成電線纏身，致使使用者不便於活動的問題。我們 的系統排除位置追蹤器要靠電線傳送感測器數據到電腦的需要，改成只要 在人體上安置紅外線發光二極體，經電視攝影機拍攝二維影像，即可以利 用電腦視覺原理，推導出三維肢體的位置與方向，如同用人眼觀察到使用 者身體的移動，而不必再由紅外線感測器拉電線傳數據給電腦。文中探討 估算三維線及平面方程式的基本理論。由於影像雜訊會造成估算誤差，故 我們先採用cross ratio 不變量原理去消除雜訊，再進行定位估算。實驗 證實有經過雜訊消除的定位估算，穩定而精確。 接著 ，我們利用3D Studio 軟體工具建立木偶各部位的三維幾何模型。有了木 偶的幾何模型後，我們可以透過VR編導軟體World Tool Kit 編排木偶的 肢體 依照設定的運動參數做運動。若木偶肢體的運動參數所設定的值， 與前述電腦視覺技術所決定的真人肢體運動參數值相同，則我們可以透過 真人的實際運動來指揮木偶的動作。如此一來，這種真人的肢體運動可以 接合虛擬實境的應用環境，做各種與人體運動有關的VR系統的製作。 In the thesis we attempt to remove the cable connecting the position tracker worn by the user to the computer. One of the problems of today(s tracking technology is that most of the position trackers need a cable to communicate with the computer. This cable will constrain the range of the motion of the user. Instead of trasmitting the data from the sensor placed on the user to the computer through the cable, we put the infared LEDs on the human body and the sensor on the computer side in such a way that there is no cable connecting the human body to the computer. By using the computer vision techniques, the 3D position and orientation of the human body can be estimated from the image photographed by computer, just like the human eyes perceive the 3D motion of an object. In this study, we formulate the equations of estimating the 3D line and plane equations. Because significant errors may arise due to the image noise, the cross ratio invariance is used to handle the image noise before calculating the 3D position. As the experiments show, the estimation results are robust and accurate with noise removal. On the other hand, we shall construct the 3D geometry model of a virtual human by using the 3D Studio tool. Then we attempt to use theVR authoring tool called World Tool Kit (WTK) to program the virtual human to follow the body motion of a real person.The body motion parameters of the real person are obtained from the above-mentioned vision-based tracker. In this way, we can guide the virtual human to act as the real person does. Thus, many virtual reality applications for the human sports can be developed by combining the real person motion and the virtual reality sports environment.