手持型裝置中三維繪圖之彩現架構與模型傳輸

Abstract

隨著三維繪圖應用的普及，在手持型裝置中實現三維繪圖彩現能力，將會是未來的潮流。因手持型裝置中所配備的運算硬體資源較少，而且所能使用到的網路頻寬也相對較低。因此我們需研發具運算效能的彩現架構與模型傳輸方式，以提供足以滿足各項三維繪圖應用需求的能力。在此論文中，我們深入探討三維繪圖彩現過程中資料流的特性，並考慮手持型裝置所特有的狀況，而研發出兩類具有運算效能的三維繪圖彩現架構，以及一種在網路環境下減少模型初始傳輸時的停滯感的方式。 我們研發的第一種三維繪圖彩現架構，根基於延遲光源照明運算以及索引式彩現技術。這架構可以避免在隱藏的多邊型與圖點上做多餘的運算，因而可以提高硬體的使用效能。經過模擬後，發現此架構在平塗著色法與高氏著色法中可以消去 10% ~ 70% 的光源照明運算，在馮氏著色法中可消去30% ~ 95%。除此之外，此架構可進一步減少彩現架構前級的幾何轉換量，若由 CPU 執行，約可消去 56.1% ~ 78.4% 的運算。而且此架構更可以將彩現運算的部分工作後延，如著色與紋理貼圖的運算均可以移至掃瞄速率區域。由於在彩現運算與掃瞄輸出兩項工作時間中取得平衡，因而可進一步提升整體彩現速率。我們研發的第二種三維繪圖彩現架構，利用三重佇列架構，實現延遲光源照明運算。此架構一樣能因延遲光源照明運算來減少不必要的光源運算。相較於傳統單一佇列的架構，此三重佇列的運算更可以有效的加速彩現運算。根據 cycle-true 模擬，此架構可以將彩現所需的時間，縮短至原先的 52.9%。在模型傳輸方面，我們研發了進場漸進式模型。此傳輸模式可以在模型尚未傳輸完之前，網路的 client端即可以開始繪出三維畫面，而減少使用者等待的時間。模擬結果顯示，在特定視角下，約只需接收到 24.2%~54% 的部分模型檔案，client端即可以開始繪製畫面，因而減少減少模型初始繪出前傳輸時的停滯感。 我們研發的兩類三維繪圖彩現架構，以及模型傳輸方式，可以在手持型裝置中，有效的改善三維彩現的運算效能與模型傳輸的使用者觀感。因此，我們在此論文中所提出架構與方法，將可有助於建立一個普及的行動三維彩現平台，以支援各項互動式的三維繪圖應用。

3-D graphics applications have played an important role in multimedia systems. With the progressive of VLSI technologies, it is a future trend to equip handheld devices with 3-D graphics abilities. Because of relative low computing power and narrow network bandwidth in handheld devices, we need to explore efficient rendering architectures and model transmission methods to offer enough performance for various applications. In this dissertation, we investigate on the nature of data flows in 3-D rendering, and propose efficient methods to improve the performance of 3-D graphics in handheld devices, including two sorts of rendering architectures and one model transmission method. The first rendering architecture bases on our deferred lighting and index rendering techniques. It can reduce the unnecessary operations on invisible polygons and pixels to improve hardware utilization. By simulation, the result shows that index rendering can eliminate 10% ~ 70% lighting operations in flat and Gouraud shading, and 30% ~ 95% in Phong shading. Besides, this architecture can reduce equivalent CPU cost on geometry processing into 56.1% ~ 78.4% compared with traditional architectures. Furthermore, this architecture can move parts of rendering jobs (shading and texture mapping) into scan-out rate area to speeded up rendering, because of the balance on computing load. The second architecture is based on our deferred lighting technique and triple queue structure. The architecture benefits from deferred lighting to reduce lighting operations. Besides, since the queue can avoid pipeline stall to speedup rendering, the triple queue structure has better performance than the traditional single queue structure. By cycle-true simulation, the triple queue structure can reduce rendering cycles into 52.9%. For model transmission, we propose entrance progressive model (EPM). With the EPM, the handheld devices can render image in the halfway of model transmission, and therefore the initiation time is shortened. The simulation results show that the models can be rendered from a specific viewpoint when only 24.2%~54% of EPM files are transmitted, and hence the EPM can reduce the feeling of waiting in model transmission. In the dissertation, we proposed two architectures and one transmission method to improve the efficiency of rendering and model transmission in handheld devices. Therefore, the proposed techniques can be utilized to construct mobile rendering platforms to support widespread interactive 3-D graphics applications.