電腦動畫中高效率骨架動作壓縮技術之研究

Abstract

在人物動畫合成中，動態捕捉技術已成為獲得擬真動作最直接也有效的方法。隨著與日遽增的動作資料，如何有效的儲存與壓縮動作資料，變成了迫切需求之多媒體處理技術。目前之骨架動作壓縮技術，雖然獲得不錯的壓縮比率。但是對於人物細微轉折等特性資料較難保存，此外，近來提出較高壓縮倍數的演算法，需利用到數值最佳化逼近目標函數以解碼，對於即時應用較不適用。 本計畫以兩年的時間，研發新的動作分段與壓縮技術。我們預期將更進一步的利用動作資料中時間與空間的相依性，並且更精準地分割重複性動作，以進一步提高壓縮率，並減少近似動作之失真率。並且達到超過即時運算速度需求的壓縮與解壓縮執行效率。 除了針對單一動作資料串，本計畫並發展對於動作資料庫之壓縮技術以及利用多執行緒平行處理。當用於即時互動應用中（如三維遊戲），可大幅增進記憶體的使用彈性，並且提升人物動作合成之擬真與變化性。

In computer animation, motion capture has become the most straightforward and the most effective approach to acquire realistic motions. However, after more and more motion data are acquired, how to efficiently store or compress the motion data becomes an urgent issue. Recently, several compression methods are proposed for skeletal motions and are of significant compression ratios. Nevertheless, detailed transitions or distinct postures may not be preserved with their spline approximation. Besides, the methods with significantly high compression ratio use run-time optimization and are not adequate to current real-time applications. In the proposed project, we plan to develop a novel motion compression method in two years. We’ll further utilize the spatial and temporal coherence to improve the efficiency of approximation. Besides, we also plan to research on segmentation of short-term repetitive motions and motion comparisons. The proposed method will further improve the compression ratios and preserve the motion details as well. The proposed compression method, through multi-threading programming, can be applied to motion database compression with high computational performance. When applying to real-time interactive applications (e.g. 3D games), the flexibility of memory usages will dramatically increase.