- 學位論文
以OpenCL技術加速三維模型成像應用程式
Case Study: Accelerating 3D Model Reconstruction with OpenCL Techniques
若您是本文的作者，可授權文章由華藝線上圖書館中協助推廣。
摘要
現今3D 科技的應用越來越廣泛了，如何建立3D模型是一件重要的事情。以往要重建一個3D模型，要依賴硬體的配合或是有很多限制。我們在這篇paper中提出了一個比較少限制的方法並且將其移植到Android環境。然而這個應用需要大量的運算，在InFocus M320上要建立一個薑餅人的3D模型大約要200秒。我們透過OpenCL去平行在Image-Based Visual Hull上的運算來加速程式的運行。在kernel的部分藉由8核心的CPU去做運算 有5.5的speedup，整體而言有2.3的speedup。另外我們找出了在我們定義的Image-Based Visual Hull範圍中最花時間的部分，透過增加一個額外的index table來大幅降低尋找資料以及運算的時間。透過這樣的改善我們將這個kernel的運算部分從26秒左右降低到只要大約0.1秒
並列摘要
3D technology is more and more popular in the world. Therefore, how to construct a 3D model has become important nowadays. In the past, it is restricted with hardware or has a lot of constraints. Comparing with the methods, we propose a 3D model reconstruction flow with less constraints and porting it to Android environment. However, this method involves large computation. It costs about 200 seconds to construct a gingerman model on InFocus M320. We parallel the computation on Image-Based Visual Hull by OpenCL. By using 8 cores of CPU, the kernel achieves a speedup of 5.5 over the sequential version. On the other hand, the complete flow achieves a speedup of 2.3 over the sequential version. Besides, we further point out which component costs too much time, and also improve the computation by adding an additional index table. Through this progress, the time of computation of the kernel can be reduced from about 26 seconds to 0.1 seconds.
參考文獻
[2] khronos, "The open standard for parallel programming of heterogeneous systems," https://www.khronos.org/opencl/ .
[6] khronos, "The opencl specication version 1.2,"
[Online]. Available: http://doi.acm.org/10.1145/383259.383266
face laser scanner for orthodontic applications," American
Journal of Orthodontics and Dentofacial Orthopedics, vol.