近年來電腦計算能力的進步使得3D虛擬環境得到廣泛的應用。本研究希望能在虛擬環境中結合人體動畫和音樂的特色，以人體動畫來詮釋音樂。我們希望能設計一個智慧型的人體動作產生器，賦予虛擬人物表達音樂特徵的能力，讓動作會因為“聽到”不同的音樂而有所不同。基於人類聽覺的短暫性，系統會自動抓取音樂特徵後將音樂切割成多個片段、對每一片段獨立規劃動作並產生動畫。過去動畫與音樂相關的研究中，許多生成的動作都經由修改或重組運動資料庫中的動作。本研究分析音樂和動作之間的關係，使用程序式動畫產生法自動產生多變且適當的詮釋動作。實驗顯示本系統能通用於LOA1人體模型和MIDI音樂；此外，透過調整系統中的參數，我們能產生不同風格的動畫，以符合不同使用者偏好和不同音樂曲風的特色。 / In recent years, the improvement of computing ability has contributed to the　wide application of 3D virtual environment. In the thesis, we propose to combine character animation with music for music interpretation in 3D virtual environment. The system proposed in the thesis is an intelligent avatar motion generator, which generates expressive motions according to music features. The system can extract music features from input music data, segment a music into several music segments, and then plan avatar animation. In the literature, much music-related animation research uses reconstruction and modification of existing motion to compose new animations. In this work, we analyze the relationship between music and motions, and then use procedural animation to automatically generate applicable and variable motions to interpret music. Our experiments show that the system can accept LOA1 models and midi as inputs in general, and generate appropriate expressive motions by modifying parameters according to users’ preference or music style.

在3D虛擬環境中，使用者透過對化身的控制進行場景瀏覽等各種活動。所謂智慧型化身之控制，是指在化身控制系統中，融入一些人工智慧的方法，協助使用者更有效率的控制化身的運動。本論文對此議題的研究範疇包含兩部分：改進第一人稱控制化身的智慧型介面及將智慧型控制的概念延伸到第三人稱的控制方式上。在以第一人稱控制化身的部分，我們在一個具備虛擬力場輔助的控制系統上，加入適性化的機制，以產生能適應使用者操控習慣的虛擬力場。另外，在一個使用街圖運動計畫法輔助控制的智慧型介面上，我們以虛擬鏈結的觀念改善該方法可能造成使用者繞路的不自然現象。在第三人稱的化身控制上，我們以「組態-時間」街圖的新觀念設計運動計畫演算法，並加上讓運動計畫與控制播放兩個程序協調並行的機制，使系統能在符合即時性的效能下，自動為化身搭配能夠適時閃避障礙物的上半身肢體運動。我們相信此類智慧型控制介面，可增加化身在虛擬環境中控制的便利性，並提高化身在動作呈現上的擬真度。 / In a 3D virtual environment, a user takes part in various activities such as walk-through a scene by controlling the avatar representing the user. Intelligent avatar control means a control mechanism using AI techniques to improve user naviga-tion efficiency. This thesis reports research on two aspects of intelligent avatar control: we attempt to improve the efficiency of the first-person intelligent ava-tar control interfaces and extend the intelligent interface concept to a third-person control mechanism. First, we improve a force-field based intelligent interface by adopting an adaptive mechanism that can generate a tailored force field according to user control behavior. In addition, in a motion-planning based assisting mechanism, we have introduced the concept of virtual link to improve the quality of the generated avatar path. Second, we use the concept of C-T space roadmap to design a new motion planning algorithm to generate collision-free motion for a third-person-controlled avatar. We have designed a mechanism to allow motion generation and execution to happen in parallel and collision-free motions for the upper body of the avatar can be generated in real time. We be-lieve that this type of intelligent control interface can enhance the efficiency of avatar interface control and improve the realism of the generated avatar motions.