電腦動畫內容之分析與理解

Computer Animation Content Analysis and Comprehension

黃科森(Ke-Sen Huang)

指導教授 ： 楊熙年

國立清華大學/電機資訊學院/資訊工程學系/博士(2010年)

https://doi.org/10.6843/NTHU.2010.00438

摘要

動畫資料之製作成本甚高，因此如何有效的重複使用既有動畫資料是目前重要研究課題之一。在重複使用既有的資料之前，首要任務即是先了解其內容。此外，由不同角度觀察同一個資料往往也會得到不同的內容資訊。因此，我們提出萃取簡要且具代表性的動畫資訊來協助使用者能快速對繁瑣的動畫資料有更明確清楚地瞭解。在本論文中，我們首先將各種動畫特徵歸納為兩類：代表屬性 (representative attributes) 以及隱性知識 (implicit knowledge)。代表屬性為動畫的視覺特徵而隱性知識則是潛在的關係。

在本篇論文的第一部份，我們針對代表屬性中的關鍵畫格 (keyframe) 提出兩種萃取方法，分別為Key Probe和FrameRank。第一種方法我們將萃取關鍵畫格視為一個矩陣分解的問題 (matrix factorization problem) 並採用最小平方法(least-squares optimization) 來求得結果。此外，本論文基於動畫畫格的屬性提出第二個稱之為FrameRank的方法。此方法利用畫格屬性來評量畫格重要性，並且依據得到的畫格重要性來萃取關鍵畫格。上述兩種方法可適用於各種動畫資料，包含了多物件動畫 (multiple objects)、剛性 (rigid-body) 和軟性 (soft-body) 的動畫物件以及會隨著時間改變結構 (time-varying structure) 的動畫物件 (例如煙火)。

本論文第二部份則是提出發掘人體動作隱性知識的方法。此方法的主要目的為展示不同動作中人體各部位的關係。其關鍵點在於針對人體動作提出一個符號表示法 (symbolic encoding scheme)。符號表示法可以有效的將人體動作表示為動作符號 (verbal symbol) 且包含動作的時間和空間上變化關係。本論文根據動作符號提出兩種視覺表示法 (visual representation) 來視覺化 (visualize) 動作資料。第一種表示法稱之為動作植物 (motion plant)。可以用來呈現動作中不同身體部位的動作先後關係。第二種表示法稱之為動作圖像 (motion icon)。此方式採用統計的觀點來展示動作中隱含的資訊。根據實驗顯示，本論文所提出之兩種視覺化方法不僅可以發掘和呈現動作中隱性的資訊，還可以用來了解人體動作裡各個身體部位的動作協調程度 (motion synergy)。

關鍵字

動畫理解 ； 動畫分析 ； 關鍵畫格 ； 動畫摘要 ； 視覺化分析

並列摘要

Prior to reusing existing data, the most fundamental task is to understand the essence of the data. Furthermore, observing given data from distinct perspectives will provide different insights. Therefore, we seek to extract informative knowledge of animations for obtaining better clarity and usability. In this dissertation, we classify various animation features into two kinds of characteristics, representative attributes and implicit knowledge. Representative attributes denote the elements that can convey visual characteristics of animations, and implicit knowledge refers to those correlations and rules which potentially exist but are not presently evident.

To obtain representative attributes, we present two animation summarization methods for extracting the keyframes of animations. The first approach is a constraint-based technique, called Key Probe, which casts the keyframe extraction problem as a constrained matrix factorization problem and solves the problem based on the least-squares optimization technique. The second method, called FrameRank, considers the frame dependent properties to evaluate frame importance and extract salient frames. Experiments with various types of animation examples show that the proposed method produces satisfying results. This method improves upon previous work in that it can deal with multiple objects, handle both rigid-body and soft-body animations, and cope with more complex animations that have a time-varying structure (e.g., fractures and fireworks).

To discover the implicit knowledge of human motions, we present a novel visual analysis approach for revealing the relation between individual motions of different body portions. The kernel of our analysis method is a symbolic encoding scheme that represents the motion patterns with verbal symbols, effectively characterizing the spatio-temporal variation of the given motion. Two new visual representations are built on the basis of the verbal symbols: a motion plant that sequentially provides a detailed characterization of the motion, and a motion icon that statistically summarizes the comparisons between verbal streams of body portions. Experimental results show that our approach not only reveals the implicit relationships of human motions, but also favors comparative visualization of human motions in the context of the motion synergy study.

並列關鍵字

Animation Understanding ； Animation Analysis ； Keyframe ； Animation Summarization ； Visual Analysis

參考文獻

Alexa, M. & Muller, W. (2000). Representing animations by principal components. Computer Graphics Forum, 19, 411-418.

Arikan, O. & Forsyth, D.A. (2002). Interactive motion generation from examples. ACM Transactions on Graphics, 21, 483-490.

Arikan, O., Forsyth, D.A. & O'Brien, J.F. (2003). Motion synthesis from annotations. ACM Transactions on Graphics, 22, 402-408.

Aubry, M., Julliard, F. & Gibet, S. (2009). Modeling joint synergies to synthesize realistic movements. In Proceedings of The 8th International Gesture Workshop.

Baak, A., Mueller, M. & Seidel, H.P. (2008). An effcient algorithm for keyframe-based motion retrieval in the presence of temporal deformations. In MIR '08: Proceeding of the 1st ACM international conference on Multimedia information retrieval , 451-458.