以視覺為基礎的角色動畫系統

Abstract

本篇論文提出了一種在有限空間下用上半身動作驅動的表演式動畫系統。我們利用無標記式電腦視覺手臂追蹤系統得到使用者的上半身動作資訊。無標記式的追蹤系統通常雜訊較多，因此首先使用濾波器去除雜訊。去除雜訊之後，使用者上半身的動作資訊可以用來直接控制動畫人物的上半身，也能經過動作辨識觸發預先設定好的動作並套用在動畫人物上 在本系統中，動畫人物的控制分為上半身及下半身兩個部份。上半身的動作可以由使用者上半身的動作去除雜訊後直接控制，也可以是這些動作經由動作辨識觸發的預錄動作，或是前兩者的混合。下半身的動作則都是動作辨識觸發的預錄動作。上下半身的動作合併起來有可能是違反物理法則的，我們提出了一個兩層式的動作整合框架來解決這個問題。在每一層中使用卡爾曼濾波器將動作變得平順，接著根據物理限制去修改動作。透過這個動作整合框架處理後，合併後的動作可以兼顧平順以及符合物理法則。

This thesis proposes a performance animation system controlled by upper body motion under a restricted workspace environment. The upper body motion are obtained from a markerless vision-based arm tracking system. Since 3D pose information derived from vision is often coarse and inexact, we apply a filter to reduce noise. After noise reduction, the motion signal can be used to directly control the animation of the upper body of the avatar, or be recognized as motion command to assign motion to the avatar. The motion control of our system is divided into two parts, the upper body and the lower body. The upper body motions of avatar can be controlled by control signal from filtered vision input, or they can be motion capture data selected by command signal from motion recognition, or they can be the mixture of two. The motions of lower body are motion capture data automatically selected by motion recognition of filtered vision input. The combined motion of the upper and the lower body may not be physically plausible. To solve this problem, we propose a two level motion integration framework. In each level, we use Kalman filters to smooth the combined motion and then revise it by imposing constraints. The combined motions are processed by the framework to ensure that they are not only smooth but also physically plausible.