本文關(guān)鍵詞： GPGPU MIC 目標(biāo)分類(lèi) 自適應(yīng)增強(qiáng)學(xué)習(xí) 車(chē)輛識(shí)別　出處：《華南理工大學(xué)》2015年碩士論文　論文類(lèi)型：學(xué)位論文

【摘要】：目標(biāo)分類(lèi)處于智能視頻監(jiān)控分析系統(tǒng)中的關(guān)鍵環(huán)節(jié),因?yàn)楫?dāng)我們需要對(duì)視頻序列中的運(yùn)動(dòng)目標(biāo)進(jìn)行監(jiān)測(cè)時(shí),要先檢測(cè)、再分類(lèi),最后才是運(yùn)動(dòng)軌跡分析和理解。其中,Adaboost(Adaptive Boosting,自適應(yīng)增強(qiáng)學(xué)習(xí))分類(lèi)算法應(yīng)用最為廣泛,其核心思想是針對(duì)同一個(gè)訓(xùn)練集訓(xùn)練不同的弱分類(lèi)器,然后把這些弱分類(lèi)器集合起來(lái),構(gòu)成一個(gè)更強(qiáng)的最終分類(lèi)器。但是要想獲得性能較好的Adaboost分類(lèi)器,往往需要花費(fèi)大量的時(shí)間在樣本訓(xùn)練上。而且訓(xùn)練算法需要占用較大的內(nèi)存空間,普通家用電腦由于自身計(jì)算能力方面能力的限制,Adaboost算法在普通電腦上難以開(kāi)展。為了降低傳統(tǒng)Adaboost算法訓(xùn)練時(shí)間,本文結(jié)合當(dāng)下主流的多核協(xié)處理器——MIC和GPGPU,針對(duì)Adaboost特點(diǎn)開(kāi)展了相關(guān)的并行優(yōu)化工作:1、對(duì)傳統(tǒng)Adaboost算法進(jìn)行熱點(diǎn)分析,發(fā)現(xiàn)Adaboost算法在訓(xùn)練弱分類(lèi)器時(shí),90%以上的耗時(shí)集中在特征值計(jì)算及排序上面;2、結(jié)合協(xié)處理器不同的硬件架構(gòu)及編程風(fēng)格,在GPGPU平臺(tái)上采用并行雙調(diào)排序方式優(yōu)化排序,改變了原有數(shù)據(jù)的存儲(chǔ)方式,減少隨機(jī)訪存的時(shí)間,提高了樣本訓(xùn)練速度。此外,本文還針對(duì)MIC平臺(tái)開(kāi)展了相應(yīng)的并行優(yōu)化實(shí)驗(yàn),對(duì)熱度相對(duì)集中的函數(shù)使用MPI/OpenMP等并行編程工具進(jìn)行并行化,且細(xì)分為粗粒度并行以及細(xì)粒度并行兩種優(yōu)化策略。在樣本數(shù)為25600,樣本大小為18*18,MIC與GPGPU分別獲得3.8和7.2加速比。實(shí)驗(yàn)表明,GPGPU在處理圖像數(shù)據(jù)方面更為出色。為了提高算法分類(lèi)準(zhǔn)確率,在樣本選取方面,本文提出了新型的樣本采集方式,在對(duì)車(chē)輛進(jìn)行識(shí)別時(shí),極大提高了原算法的檢測(cè)精度。此外,針對(duì)目標(biāo)分類(lèi)檢測(cè)過(guò)程中速度較慢的問(wèn)題,提出了并行優(yōu)化方案。
[Abstract]:Target classification is a key step in the intelligent video surveillance and analysis system, because when we need to monitor the moving targets in video sequences, we must first detect and then classify. Finally, the analysis and understanding of motion trajectory. Among them, Adaboostan Adaptive boost (Adaptive Enhancement Learning) classification algorithm is the most widely used. Its core idea is to train different weak classifiers for the same training set, and then set these weak classifiers together. To form a stronger final classifier, but to obtain a better performance Adaboost classifier. It often takes a lot of time to train samples, and the training algorithm needs to occupy a large amount of memory space. The common home computer is limited by its own computing ability. In order to reduce the training time of traditional Adaboost algorithm, this paper combines the current mainstream multi-core coprocessor, MIC and GPGPU. According to the characteristics of Adaboost, the parallel optimization work: 1 is carried out. The focus of traditional Adaboost algorithm is analyzed, and it is found that Adaboost algorithm is training weak classifier. The time consuming over 90% is concentrated on the calculation and ranking of eigenvalues; 2. Combined with the different hardware architecture and programming style of the coprocessor, the parallel double-tone sorting method is adopted to optimize the sorting on the GPGPU platform, which changes the storage mode of the original data and reduces the time of random memory access. The training speed of the sample is improved. In addition, the parallel optimization experiments are carried out for the MIC platform. The functions with relatively concentrated heat are parallelized by parallel programming tools such as MPI/OpenMP and subdivided into coarse-grained parallelism and fine-grained parallelism. The number of samples is 25600. The sample size is 18 ~ (18) mics and GPGPU is 3. 8 and 7. 2 speedup ratio, respectively. The experiment shows that. In order to improve the classification accuracy of the algorithm, in the aspect of sample selection, this paper proposes a new way to collect samples, which is used to identify vehicles. The detection accuracy of the original algorithm is greatly improved. In addition, a parallel optimization scheme is proposed to solve the problem of slow speed in the process of target classification detection.
【學(xué)位授予單位】：華南理工大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2015
【分類(lèi)號(hào)】：TN948.6;TP301.6

【參考文獻(xiàn)】

相關(guān)期刊論文 前9條

1 孫偉平;向杰;陳加忠;余勝生;;基于GPU的粒子濾波并行算法[J];華中科技大學(xué)學(xué)報(bào)(自然科學(xué)版);2011年05期

2 程峰;李德華;;基于CUDA的Adaboost算法并行實(shí)現(xiàn)[J];計(jì)算機(jī)工程與科學(xué);2011年02期

3 郭靜;陳慶奎;;基于CUDA的快速圖像壓縮[J];計(jì)算機(jī)工程與設(shè)計(jì);2010年14期

4 左登宇;董蘭芳;宋波;;Adaboost人臉檢測(cè)方法及其并行實(shí)現(xiàn)[J];計(jì)算機(jī)仿真;2010年06期

5 趙雪竹;王秀;朱學(xué)峰;;基于Adaboost算法的人眼檢測(cè)中樣本選擇研究[J];計(jì)算機(jī)技術(shù)與發(fā)展;2010年02期

6 左顥睿;張啟衡;徐勇;趙汝進(jìn);;基于GPU的快速Sobel邊緣檢測(cè)算法[J];光電工程;2009年01期

7 明安龍;馬華東;;多攝像機(jī)之間基于區(qū)域SIFT描述子的目標(biāo)匹配[J];計(jì)算機(jī)學(xué)報(bào);2008年04期

8 王海川,張立明;一種新的Adaboost快速訓(xùn)練算法[J];復(fù)旦學(xué)報(bào)(自然科學(xué)版);2004年01期

9 王亮,胡衛(wèi)明,譚鐵牛;人運(yùn)動(dòng)的視覺(jué)分析綜述[J];計(jì)算機(jī)學(xué)報(bào);2002年03期

相關(guān)碩士學(xué)位論文 前4條

1 李紅艷;交通監(jiān)控系統(tǒng)中的運(yùn)動(dòng)目標(biāo)檢測(cè)與分類(lèi)[D];太原科技大學(xué);2012年

2 朱建清;智能視頻監(jiān)控系統(tǒng)中的人臉檢測(cè)和人臉跟蹤技術(shù)的研究[D];華僑大學(xué);2012年

3 陳翰波;基于部分集分類(lèi)器和并行計(jì)算的人臉檢測(cè)訓(xùn)練[D];中南大學(xué);2011年

4 劉麗麗;基于形狀特征的運(yùn)動(dòng)目標(biāo)分類(lèi)方法研究[D];湖南大學(xué);2006年

，

本文編號(hào)：1477398