動態網格之姿勢相依多層精細度模型

Abstract

三維動畫在現在已經非常普遍的被使用在電影和遊戲之中。因為我們通常是即時地撥放在遊戲中使用的三維動畫，美術人員所製作的複雜模型通常會被簡化為較少多邊形數目的模型。雖然已有許多為了靜態模型所提出的化簡方法，只有很少數方法能處理可動模型。而現存處理可動模型的方法不是太花計算資源，就是無法產生令人滿意的結果。在這篇論文中，我們提出一個有效率的姿勢相依多層精細度模型的架構來處理可動模型。我們的方法包含了兩個階段。第一個前處理階段執行所有須大量計算的步驟，並以共同的頂點架構為每個關鍵畫面產生一個化簡過後的模型。第二個步驟以混合連接關係和內插幾何位置來產生在兩個連續的關鍵畫面之間的網格資料。我們觀察到所提出的發法能夠有效率地且平滑地在兩個連續的姿勢相依模型間轉換，且只須很少的額外資料空間。

3D animation is now very popular in the movies and games. Since 3D animation in game is usually performed in real-time, complex models created by artist are often simplified to models of low polygon count. While many simplification methods have been proposed for static models, very few methods are available for the deformable meshes. Current simplification methods for deformable meshes are either computationally expensive or incapable of producing satisfactory results. In this thesis, we propose an efficient pose-dependent level of detail frame work for deformable meshes. Our approach consists of two stages. The first stage preprocesses all costly computation steps and for each frame generates a simplified mesh using a common vertex hierarchy. The second stage generates in-between meshes between two consecutive key frames via connectivity blending and geometry interpolation. We have observed that the proposed method is able to efficiently generate smooth transition between two pose-dependent meshes, while requiring very little extra data storage.