基於CUDA技術之視訊強化演算法的最佳化

視訊強化主要是藉由處理亮度、飽和度或雜訊等特徵值將陰暗、濃霧或失焦視訊調整為較清晰的畫面。很多視訊分析應用的前置的工作都需要用到視訊分析的技術，但是相對較大的運算量卻是視訊強化在應用上的一大問題。通用圖形處理器(GPGPU)是一種使用圖形處理器(GPU)來取代傳統的中央處理器以進行平行運算的技術。依據這個概念，NVIDIA推出了CUDA這個通用計算的架構（具備新的平行程式設計模型和指令集架構）來更有效率地解決許多複雜的計算問題。在本論文中，我們概述了視訊強化演算法及CUDA的程式設計模型，並且實作了一個視訊強化演算法的最佳化來說明從傳統的中央處理器運算轉換成平行的圖形處理器運算所能得到的加速效果。由實作的結果可以得知，藉由CUDA來對視訊強化演算法作最佳化後，對於原來的CPU程式或是尚未最佳化前的GPU程式，都可達到數倍以上的加速效果。

關鍵字

視訊強化；平行運算；統一計算架構；圖形處理器

並列摘要

Video Enhancement is to adjust the dark, fog or out of focus to be a more clear video mainly by processing the video brightness, saturation or noise and other features of value. A lot of video analysis applications need to use video analysis technique as a pre-work, but the relatively large amount of computation is to strengthen the application of video on a big problem. General-purpose computing on graphics processing units (GPGPU) is a technique of using graphics processing units (GPU) as data-parallel computing devices instead of performing computations in traditional CPUs. According to this concept, nVIDIA introduced a General-Purpose Parallel Computing Architecture (CUDA) as a general-purpose computing architecture (with a new parallel programming model and instruction set architecture) to solve many complex computational problems in a more efficient way. In this paper, we summarize a video enhancement algorithm and the CUDA programming, and implement a video enhancement algorithm optimization to illustrate the speedup from traditional CPU computations to parallel GPU computations. The implemental result illustrates that more than several times acceleration effect can be achieved compared to original CPU program or non-optimized GPU program by using CUDA to optimize the video enhancement algorithm.

並列關鍵字

Video Enhancement ； Parallel Computation ； Compute Unified Device Architecture ； CUDA ； Graphics Processing Units ； GPU

國際替代計量

基於CUDA技術之視訊強化演算法的最佳化

全文下載

主題瀏覽