摘要
本篇論文提出一個基於球型座標的影像轉換系統，經由結合OpenGL的三維呈現以及OpenCL的平行運算能力，藉此達到一個可互動式的3D虛擬場景即時影像。由於日常生活中一般相機所拍攝的單張照片視野廣度很有限，所以通常會藉由影像縫合(Image Stitching)的技術，將多張場景相鄰的照片接合，藉以得到一張更大視野的影像，甚至可以達到360度的環場效果。在此我們使用OpenCL平行加速運算技術，將多張影像即時轉換成一張全景貼圖樣式，並且以3D模型方式呈現，相較於以CPU方式處理，當處理影像越多越能凸顯平行運算的效能優勢，實驗中顯示在使用NVIDIA 9600GT顯示卡同時處理八張640*480影像時，計算效能能夠達到90倍的提升，並且能夠與使用者即時互動。
並列摘要
In this paper, we implement a video-based spherical coordinate conversion system, through the combination of OpenGL which have the 3D stereo effect and OpenCL massive parallel computing power to achieve an interactive real-time 3D virtual scene image. Due to the general everyday life single photo taken with the camera view is very limited in scope, in order to get a larger image view even can reach 360 degrees surround effect we usually use image stitching (Image Stitching) technology rather than pictures of the scene adjacent joints. Here we use OpenCL parallel computing techniques to immediately convert multiple images into a panorama texture and present it as a 3D model, compared to traditional CPU approach, we can tell that the more images to process the better efficiency of parallel computing we can get. During the experiment, it shows that when we use NVIDIA 9600GT to process eight 640*480 images, OpenCL can at most enhance efficiency up to ninety times to be able to achieve real-time interaction.