改進可視度偵測以用於手持裝置上之軟體三維繪圖引擎

Abstract

在本篇論文中，我們主要的目標就是改善傳統三維繪圖引擎的運算效率，針對其演算法上的瓶頸做修正。我們針對傳統三維繪圖引擎的每個流程階段，提出新的演算法來對運算效率或是繪圖品質做改進。 幾個主要且創新的演算法包括了：1．可視度偵測演算法，大量的減少了z-buffer所需要用的讀取與寫入的運算，同時也減少了繪圖引擎中所需要做的三角形頂點的位置轉換；2•一個新的z-buffer架構，可以減少一半的清除buffer運算；一個新的計算三維模型光影的流程，效果能夠逼近Phong的效果，但是運算複雜度接近Gouraud。 利用模擬器以及手持裝置的測試，整體的運算效率在經過這些演算法的改進之後，有了大幅度的提升。

The limited display resolution,small amount of memory, slow memory access rate and narrow bus bandwidth are critical issues that make the rendering performance and quality undesirable. This thesis will mainly focus on improving the performance bottleneck of the traditional graphics pipeline. In addition, in each stage, we present some new algorithms to either speedup the rendering performance or the rendering quality. There are some key innovations included in this thesis : I) The cone-based visibility-test algorithm, which reduces a signi‾cant amount of read/write operations on Z-buffer, and diminishes the number of vertex transformation; II) The view-flipping Z-buffer scheme that avoids half of z-buffer clear operation; III) a lighting system with rendering quality closes to Phong Shading while the computation cost is almost the same as Gauroud Shading. The performance comparisons on software simulator and real devices show that these innovations greatly improve the rendering capabilities on the mobile devices.