利用局部動作查詢與組合之角色動作合成

Abstract

近年來動作捕捉技術正被廣泛的應用在電影、電玩與多媒體相關產業中。因此，如何運用這些捕捉而來有限的動作資料一直是很重要的課題。在本篇論文中，我們將提出一個簡單又直覺的局部動作查詢與合成系統，藉由此系統有效又方便的創造出看起來自然又真實但原本資料庫中沒有的動作，以擴大動作資料庫的變化性。我們的想法是，若我們找出看起來相似的局部動作，將相似的局部動作更換後的動作該會是很接近自然的。因此，我們的系統提供一個簡單的動作編輯器，讓使用者藉由幾個關鍵的姿勢創造出期望之動作雛型。再由我們的系統經由查詢的方式至資料庫中將相似的動作片段挑出。使用者可從中選出準備接合的基礎動作與替換之局部動作來組合。為了保持原局部動作的行為變化與接合的自然性，我們利用動作統計分佈的方式做調整。本論文所提出之系統將可使使用者快速建立合理之動作片段且可擴增原資料庫之動作變化性

Using motion capture data to create novel and realistic motions is an attractive topic in motion synthesis. In this thesis, we propose a simple and intuitive editing tool to assemble the different parts of character motion to create novel and natural-looking motion. Herewith the variety of the database is enriched. In our system, users can assign the partial figure of a query motion which they want to synthesize. Our system then retrieves partial figures from large data sets similar to the assigned ones. Finally, we assemble the different parts of character motion. However, simply assembling partial figures may cause unrealistic or strange motions. To solve this problem and also keep the query motion style, we use the statistic properties between the query motion and pre-assembly motion. When assembling the retrieval partial figures and adjusting the variance according to the query motion, our system can retain the style and naturalness of the synthesized motion.