【正文】
分類號(hào) TP391 密級(jí) 公開重慶郵電大學(xué)碩士學(xué)位論文論文題目 基于光流的運(yùn)動(dòng)估計(jì)與匹配方法研究(題名和副題名)英文題目 Research on the Methods of Motion Estimationand Matching Based on Optical Flow碩士研究生 李文羽 指導(dǎo)教師 胡學(xué)剛 教授 學(xué)科專業(yè) 計(jì)算機(jī)應(yīng)用技術(shù) 論文提交日期 2010年5月 論文答辯日期 論文評閱人 答辯委員會(huì)主席 2010年5月16日獨(dú) 創(chuàng) 性 聲 明本人聲明所呈交的學(xué)位論文是本人在導(dǎo)師指導(dǎo)下進(jìn)行的研究工作及取得的研究成果。據(jù)我所知,除了文中特別加以標(biāo)注和致謝的地方外,論文中不包含其他人已經(jīng)發(fā)表或撰寫過的研究成果,也不包含為獲得 重慶郵電大學(xué) 或其他教育機(jī)構(gòu)的學(xué)位或證書而使用過的材料。與我一同工作的同志對本研究所做的任何貢獻(xiàn)均已在論文中作了明確的說明并表示謝意。學(xué)位論文作者簽名: 簽字日期: 年 月 日學(xué)位論文版權(quán)使用授權(quán)書本學(xué)位論文作者完全了解 重慶郵電大學(xué) 有關(guān)保留、使用學(xué)位論文的規(guī)定,有權(quán)保留并向國家有關(guān)部門或機(jī)構(gòu)送交論文的復(fù)印件和磁盤,允許論文被查閱和借閱。本人授權(quán) 重慶郵電大學(xué) 可以將學(xué)位論文的全部或部分內(nèi)容編入有關(guān)數(shù)據(jù)庫進(jìn)行檢索,可以采用影印、縮印或掃描等復(fù)制手段保存、匯編學(xué)位論文。(保密的學(xué)位論文在解密后適用本授權(quán)書)學(xué)位論文作者簽名: 導(dǎo)師簽名:簽字日期: 年 月 日 簽字日期: 年 月 摘 要 本文主要研究了視頻圖像序列光流的運(yùn)動(dòng)估計(jì)及匹配應(yīng)用問題,它是計(jì)算機(jī)智能化的一個(gè)基本問題,也是動(dòng)態(tài)圖像分析的核心問題。圖像系列光流運(yùn)動(dòng)估計(jì)是快速而準(zhǔn)確地檢測圖像系列幀間運(yùn)動(dòng)的一類技術(shù),在經(jīng)濟(jì)和國防建設(shè)中有著廣泛的應(yīng)用前景,具有非常重要的研究意義。本文首先介紹了二十世紀(jì)八十年代以來國內(nèi)外對光流運(yùn)動(dòng)估計(jì)和匹配問題研究的各類算法,分析了光流運(yùn)動(dòng)估計(jì)及匹配方法的研究現(xiàn)狀。其次,作為理論基礎(chǔ),不但詳細(xì)推導(dǎo)了HS光流法的基本方程和數(shù)值解法,而且全面論證了Loggabor函數(shù)的性能。最后,客觀的分析了近年來光流運(yùn)動(dòng)估計(jì)和匹配算法中存在的不足之處,引出了本文研究的重點(diǎn)內(nèi)容:基于梯度優(yōu)化的多尺度視頻光流估計(jì)和基于兩步運(yùn)動(dòng)估計(jì)的系列圖像匹配算法。第一,提出一種基于梯度優(yōu)化的不同運(yùn)動(dòng)幅度視頻圖像光流估計(jì)的新算法。先用Loggabor濾波器對原視頻圖像進(jìn)行相位、尺度濾波,再用所得的特征圖像來計(jì)算時(shí)空梯度,最后根據(jù)時(shí)空梯度計(jì)算光流。該算法模型同時(shí)運(yùn)用由粗到精的圖像金字塔方法對視頻圖像分層處理。理論分析和實(shí)驗(yàn)結(jié)果表明,該算法適用于大幅度的視頻運(yùn)動(dòng)光流估計(jì),不僅能得到適合人眼視覺分辨率特性的圖像,而且使時(shí)空梯度更加優(yōu)化,光流計(jì)算更準(zhǔn)確。并且在時(shí)間復(fù)雜度上與傳統(tǒng)光流計(jì)算方法相當(dāng),在計(jì)算精度上優(yōu)于HS、段先華等人提出的算法。第二,以光流法(Optical Flow)為基礎(chǔ),結(jié)合簡化的仿射變換(Simplified Affine Transform)模型,提出了一種基于兩步運(yùn)動(dòng)估計(jì)的系列圖像匹配算法——SAT_OF算法。該算法彌補(bǔ)了傳統(tǒng)的圖像匹配方法的不足,而且使圖像匹配殘差明顯減小。此外,在保證簡化仿射變換參數(shù)正確的同時(shí),使用了更加簡單直接的解法,大大降低了計(jì)算的復(fù)雜性。實(shí)驗(yàn)表明,對存在大位移運(yùn)動(dòng)、復(fù)雜形變的圖像,該算法更有效。關(guān)鍵詞:光流;運(yùn)動(dòng)估計(jì);Loggabor濾波;金字塔方法;圖像匹配;簡化射變換AbstractThe thesis focus on the technology of motion estimation and matching of optical flow in the video image sequences, which is known as the principle field of intelligent puter and dynamic image analysis. The technology of optical flow motion estimation that can fast and accurately detect interframe motion has a wide application prospects and has a very important research significance in the areas of the economic and national defense construction.This paper first analyzes the various types of algorithms and the research situation of optical flow motion estimation and matching at home and abroad since 1980s. In the next place, as a theoretical basis, the HS’s basic equation of optical flow method and the numerical solution is not only derived in detail, but the performance of Loggabor function is fully demonstrated as well. Finally, we objectively analyze shortings of the recent optical flow motion estimation and matching algorithms, leading to the focus of the contents of this paper: multiscale optical flow estimation of the video based on gradient optimization and a Image matching algorithm based on twostep motion estimation.Firstly, a new algorithm based on gradient optimization is presented for optical flow estimation of video images with different motion range. Original video images are first transformed by using Loggabor filtering in phase and measure, and then the spatiotemporal gradient is calculated by using the images of the feature obtained, last optical flow is calculated in the light of the spatiotemporal gradient. In the meanwhile, the video images are layered and processed by the algorithm model using coarsetofine image pyramid method. Both theory and experiment show that the algorithm is applied to the video optical flow motion estimation of the significant range. The video images which are suitable for human visual characteristics of the resolution can not only be gained, but also the spatiotemporal gradient is more optimized and optical flow calculation is more accurate. Besides, The time plexity of this algorithm is equivalent to that of the traditional optical flow method, and in the accuracy of the algorithm is superior to the methods suggested by HS, Duan etc. Secondly, a twostep motion estimation algorithm is proposed to match a series of images, which is based on Optical Flow method, and bines with Simplified Affine Transform model. The algorithm pensates the disadvantage of classical image matching methods, and obviously reduces image matching error. Besides, with the use of simple and direct solution, The method can not only assures the accuracy of Simplified Affine Transform parameters but also greatly reduces putational plexity. Experiment results show that the algorithm is more effective for images with largescale motion and plex deformation.Key words: Optical flow。 Motion estimation。 Loggabor filtering。 Pyramid method。 Image matching。 Simplified affine transform目 錄摘 要…………………………………………………………………IAbstract……………………………………………………………II第一章 緒論………………………………………………………1 論文選題背景…………………………………………………………………1 光流運(yùn)動(dòng)估計(jì)研究現(xiàn)狀………………………………………………………2 基于微分技術(shù)的方法…………………………………………………2 基于頻域的方法………………………………………………………3 其它光流估計(jì)方法……………………………………………………5 小結(jié)……………………………………………………………………5 圖像系列中匹配技術(shù)研究現(xiàn)狀………………………………………………5 圖像匹配中基本的空間變換…………………………………………6 剛性匹配方法…………………………………………………………7 非剛性匹配方法………………………………………………………8 小結(jié)……………………………………………………………………9 論文主要工作………………………………………………………………10 論文結(jié)組結(jié)構(gòu)………………………………………………………………11第二章 光流運(yùn)動(dòng)估計(jì)和匹配技術(shù)基礎(chǔ)…………………………12 HS光流向量計(jì)算……………………………………………………………12 亮度恒定假設(shè)…………………………………………………………12 光流基本方程…………………………………………………………12 孔徑問題………………………………………………………………13 平滑性準(zhǔn)則……………………………………………………………13 總誤差目標(biāo)函數(shù)………………………………………………………14 時(shí)空圖像差分方法……………………………………………………15 總誤差目標(biāo)函數(shù)的最優(yōu)化方法………………………………………15 求解過程的數(shù)值計(jì)算方法……………………………………………16 loggabor濾波器……………………………………………………………17 一維loggabor函數(shù)及其特性………………………………………17 二維loggabor濾波器的構(gòu)造………………………………………19 仿射變換……………………………………………………………………20 本章小結(jié)………………………………………………………