(3.238.99.243) 您好!臺灣時間:2021/05/17 00:26
字體大小: 字級放大   字級縮小   預設字形  
回查詢結果

詳目顯示:::

我願授權國圖
: 
twitterline
研究生:王建舜
研究生(外文):Chien-Shun Wang
論文名稱:在多核心系統上物件輪廓追蹤的平行化與效能分析
論文名稱(外文):Parallelization and Performance Analysis of Contour-based Object Tracking on Multi-core Systems
指導教授:林偉林偉引用關係
學位類別:碩士
校院名稱:國立中興大學
系所名稱:資訊網路多媒體研究所
學門:電算機學門
學類:軟體發展學類
論文種類:學術論文
畢業學年度:97
語文別:中文
論文頁數:37
中文關鍵詞:多核心系統多核心平行化物件輪廓追蹤
外文關鍵詞:Multi-core systemsParallelismContour-Based Object Tracking
相關次數:
  • 被引用被引用:0
  • 點閱點閱:124
  • 評分評分:
  • 下載下載:0
  • 收藏至我的研究室書目清單書目收藏:0
  物件輪廓追蹤(Contour-based Object Tracking)方法在電腦視覺領域中是一個非常重要的任務,其已經被廣泛地應用在影片中追蹤移動的目標物件。藉由追蹤到的輪廓我們不只可以知道欲追蹤之目標物件的位置,也可以藉由分析該物件的輪廓來得知該物件的行為。因此,物件輪廓追蹤在電腦視覺領域的物件追蹤方法中是最強大的其中之一種。因此,在本論文中,我們選擇了可以同時追蹤多個物件輪廓的Yilmaz的方法,並利用平行化的方法加速其在多核心系統上的效能表現。
  在今日,多核心系統(Multi-core Systems)已經逐漸變成主流。小至個人電腦,大至超級電腦,都不難見到它的蹤跡。這意味著對於程式設計者而言,效能的提升再也不是"自動"的,必須要從傳統"單執行緒"的概念轉變為"多執行緒"。因此,平行程式設計也成為程式設計人員的一個必須面對的問題。我們可以藉由發揮多核心系統強大的運算效能來提升物件輪廓追蹤方法的效能表現。
  在本論文中,我們分析了物件輪廓追蹤方法並將其三個主要的模組平行化。利用效能分析工具找出在該模組中的熱點,也就是計算量最大的部分,並依據其來決定平行化的階段。然後將這些模組使用Intel C++ Compiler提供的OpenMP應用程式介面來進行執行緒階段平行(Thread Level Parallelism),將模組一一地平行化。之後我們再利用平行程式的分析工具-Intel VTune效能分析器與執行緒數據檔案器來觀察平行化後物件輪廓追蹤應用的平行效能。分析的結果可以幫助我們找到平行程式的效能瓶頸所在,我們也提出一些方法去解決這樣的瓶頸,使這個應用能在多核心系統上達到較好的可擴展性(Scalability)。
  Contour-based Object Tracking is a very important task in computer vision domain. It has been widely applied to tracking moving object in video sequences. With tracked contours, we can not only know where the object is but also know what the object is doing. We can analyze the tracked contour to understand its behavior. Contour object tracking is one of the most powerful methods in object tracking domain. Yilmaz’s contour-based tracking method is such a method that is capable of tracking multiple objects simultaneously. In this thesis, we are interested in accelerating Yilmaz’s method through parallelization on multi-core processors.
  Recently, the multi-core processors have become a mainstream of high-performance computer architecture. They exist in varying forms from personal computers to large-scale supercomputers. This has been an important, physical implication that performance enhancement of high-speed computing on these multi-core processors is no longer done automatically, without programmers intervention. In other words, programmers now have to face the problem of developing parallel applications with the concept of multi-thread, rather than with the conventional single-thread one. The objective of our research effort is to enhance the performance of a contour-based tracking method by making effective use of the computing power provided by the multi-core computer.
  In this thesis, we intend to parallelize the three major parts of Yilmaz''s contour-based object tracking method. We first use the Intel performance analyzer to identify hotspots that encompass the largest amount of computation. Then we parallelize them at the thread level by using C++ with OpenMP directives supported in the Intel C++ Compilers. Using Intel VTune Performance Analyzer and Intel Thread Profiler, we analyze the parallel performance of parallelized contour tracking. With the aid of analysis results, we are able to pinpoint such bottlenecks as load imbalance in some parts of contour-based tracking method, and we suggest some ways to eliminate these bottlenecks and improve the scalability of the tracking method as a consequence.
誌謝 i
摘要 ii
Abstract iii
目錄 v
圖目錄 vii
表目錄 viii
表目錄 viii
第一章 簡介 1
1.1 論文簡介 1
1.2 論文架構 2
第二章 背景知識 4
2.1 物件追蹤演算法 4
2.2 OpenMP和安達爾定律(Amdahl''s Law) 5
2.2.1 OpenMP 5
2.2.2 安達爾定律 6
2.3 多核心架構 8
2.3.1 單執行緒與同步多執行緒 9
2.3.2 對稱式多處理器架構 11
2.3.3 單晶片多處理器架構 12
2.4 相關研究 13
第三章 研究主題 14
3.1 輪廓追蹤演算法 14
3.2 輪廓追蹤演算法的分析 18
3.3 平行化策略 21
3.3.1 平行化等高集合法 21
3.3.2 平行化蓋伯濾波器 22
3.3.3 平行化艾氏核函數 23
3.4 平行效能調整 24
第四章 效能量測與實驗結果分析 27
4.1 實驗環境介紹 27
4.2 效能提升幅度 28
4.2.1 平行化等高集合法的效能分析 28
4.2.2 平行化蓋伯濾波器的效能分析 30
4.2.3 平行化艾氏核函數的效能分析 32
4.2.4 以執行緒數量為變數的實驗結果 33
第五章 結論 35
參考文獻 36
1.Yilmaz, A., O. Javed, and M. Shah, Object tracking: A survey. ACM Comput. Surv., 2006. 38(4): p. 13.
2.Dagum, L., R. Menon, and S. Inc, OpenMP: an industry standard API for shared-memory programming. IEEE Computational Science & Engineering, 1998. 5(1): p. 46-55.
3.Hill, M. and M. Marty, Amdahl''s law in the multicore era. Computer, 2008. 41(7): p. 33-38.
4.Bailey, D., et al., The NAS parallel benchmarks. International Journal of High Performance Computing Applications, 1991. 5(3): p. 63.
5.Curtis-Maury, M., et al., An evaluation of OpenMP on current and emerging Multithreaded/Multicore processors. Lecture Notes in Computer Science, 2008. 4315: p. 133.
6.Pankratius, V., et al., Software engineering for multicore systems: an experience report, in Proceedings of the 1st international workshop on Multicore software engineering. 2008, ACM: Leipzig, Germany.
7.Noronha, R. and D.K. Panda. Improving Scalability of OpenMP Applications on Multi-core Systems Using Large Page Support. in Parallel and Distributed Processing Symposium, 2007. IPDPS 2007. IEEE International. 2007.
8.Qi, Z., et al. Parallelization and Performance Analysis of Video Feature Extractions on Multi-Core Based Systems. in Parallel Processing, 2007. ICPP 2007. International Conference on. 2007.
9.Qi, Z., et al. SIFT implementation and optimization for multi-core systems. in Parallel and Distributed Processing, 2008. IPDPS 2008. IEEE International Symposium on. 2008.

10.Amit, G., et al. Scalability of Multimedia Applications on Next-Generation Processors. in Multimedia and Expo, 2006 IEEE International Conference on. 2006.
11.Yilmaz, A., L. Xin, and M. Shah, Contour-based object tracking with occlusion handling in video acquired using mobile cameras. Pattern Analysis and Machine Intelligence, IEEE Transactions on, 2004. 26(11): p. 1531-1536.
12.Inc, I., Intel VTune Performance Analyser.
13.Osher, S. and J. Sethian, Fronts propagating with curvature dependent speed: Algorithms based on Hamilton-Jacobi formulations. Journal of computational physics, 1988.
14.Intel, M., Intel Math Kernel Library.
QRCODE
 
 
 
 
 
                                                                                                                                                                                                                                                                                                                                                                                                               
第一頁 上一頁 下一頁 最後一頁 top