多層精細度模型的動態產生法

Abstract

多層次精細度模型已在虛擬實境應用被廣泛使用。其目的是在模擬中可 依視點、視線決定合適的幾何精細度，以減少需顯像的多邊形個數。傳統 的做法需事先對場景中的各物體產生數組不同精細度的模型。本論文提出 ，在模擬時，動態地產生多層次精細度模型。我們在前置處理中對整個場 景做空間切割（八元樹），記錄不同子空間（八元樹節點）之間的幾何資 訊（節點間的相連關係）。在模擬進行中，我們考慮視點或視角等觀視依 循 ( view-dependent ) 性質，以視點和節點空間之間的距離、視角（視 點和節點空間的相對位置）決定八元樹節點在視覺幾何上的重要性，再根 據需要 ( 例如維持固定frame rate的時間控制 )，選取某些在視覺幾何 效果上較重要的節點，以這些節點 ( 有效節點 ) 之間的幾何連結即時產 生不同精細度的場景模型。同時藉由frame coherence的使用，可加速節 點視覺幾何重要性的計算，並在相鄰 frame 間，只需對在兩 frames 之 間視覺效果差異大的部分有效節點做幾何精細度調整運算，以改變顯像模 型的精細度。 Level-of-detail ( LOD ) modeling has been widely employed in virtual realityapplications. In the process of sumilation，a suitable level of detail for each object isselected depending on view poing and direction. The traditional LOD modelingmethods need to generate and store a prior a set of models with differentlevel of details for all ojbects in the scene. In this thesis, we proposea method to generate the level-of-detail model dynamically during thesimulation process. In preprocesing, we subdivide the scene using octreesubdivision and record the geometry connection information betweendifferent octree nodes. During simulation, perception importance foroctree nodes is computed depending on view point and direction, and basedon that nodes with greater (smaller) perception importance are selectedto be further refined (reduced) to derive an appropriate LOD model for the scene.Morever, frame time control is implemented to obtain an constant expected frame rate.Frame coherence is used to speed up the the computation of the perception importancefor octree nodes and the refinement or reduction of scene geometry.