本文選題：光流算法　切入點：引導(dǎo)濾波　出處：《南昌航空大學(xué)》2017年碩士論文

【摘要】：光流是指物體或者三維場景在二維投影平面上像素運動的瞬時速度,其不僅攜帶了被觀察物體的運動信息,還包含了物體的三維結(jié)構(gòu)信息。因此圖像序列光流計算技術(shù)研究是圖像處理、模式識別與計算機視覺等領(lǐng)域的一個重要研究方向。21世紀(jì)以來,隨著計算機軟、硬件及顯示技術(shù)的迅猛發(fā)展,圖像序列光流計算技術(shù)已在社會生產(chǎn)及生活中的各個領(lǐng)域突顯出了越來越重要的地位和作用,研究成果被廣泛應(yīng)用于氣象分析、交通視頻監(jiān)控、無人自動駕駛系統(tǒng)、醫(yī)學(xué)圖像分割以及機器人視覺系統(tǒng)等領(lǐng)域。近年來,隨著光流計算研究的不斷深入,針對光流計算模型和優(yōu)化方法的新理論與新技術(shù)不斷涌現(xiàn),圖像序列光流計算技術(shù)在計算精度和計算效率等方面取得了突破性的進展。但是當(dāng)圖像序列中包含運動邊緣模糊、非剛性運動、亮度突變與不連續(xù)等困難場景和運動形式時,光流計算的準(zhǔn)確性和魯棒性問題還有待進一步研究。本文主要是針對圖像序列運動模糊等問題進行深入研究,通過設(shè)計基于引導(dǎo)濾波的圖像序列光流計算優(yōu)化模型,克服由復(fù)雜邊緣結(jié)構(gòu)、大位移和非剛性運動所引起的光流計算邊緣模糊等問題,以改善光流計算的準(zhǔn)確性和魯棒性。本文的主要工作包括以下幾點:1.對圖像序列光流計算技術(shù)的研究背景與研究現(xiàn)狀進行了詳細(xì)的介紹,并對現(xiàn)階段光流計算技術(shù)所面臨的主要問題與挑戰(zhàn)進行了詳細(xì)的分析。2.重點對HS、CLG-TV等經(jīng)典光流計算方法和Classic+NL、NNF等近年來涌現(xiàn)的代表性光流計算模型進行介紹,并分析各類計算方法的優(yōu)缺點及適用范圍。3.針對復(fù)雜邊緣結(jié)構(gòu)、大位移運動以及非剛性運動引起的邊緣模糊等問題,提出圖像金字塔分層細(xì)化框架下基于引導(dǎo)濾波的光流計算優(yōu)化方法。在光流計算過程中,通過引導(dǎo)濾波的線性化作用,能夠很好的保留圖像邊緣細(xì)節(jié)部分,克服了圖像邊緣模糊等問題,使得圖像邊緣達(dá)到增強和銳化的效果,提高了光流的計算精確性和魯棒性。4.分別采用Middlebury、MPI Sintel和KITTI數(shù)據(jù)庫提供的標(biāo)準(zhǔn)測試圖像集對HS、CLG-TV、Classic+NL和NNF等典型變分光流方法進行引導(dǎo)濾波優(yōu)化對比測試。實驗結(jié)果表明,本文所基于引導(dǎo)濾波的光流計算優(yōu)化模型能夠有效地提高光流計算的準(zhǔn)確性與魯棒性,尤其是當(dāng)圖像序列中包含復(fù)雜邊緣結(jié)構(gòu)和大位移運動時,能夠大幅提升光流計算的效果。
[Abstract]:Optical flow refers to the instantaneous velocity of pixels moving on the two-dimensional projection plane of an object or 3D scene. It not only carries the motion information of the object under observation, but also contains the three-dimensional structure information of the object.Therefore, the research of image sequence optical flow computing technology is an important research direction in the field of image processing, pattern recognition and computer vision. Since the 21st century, with the rapid development of computer software, hardware and display technology,Image sequence optical flow computing technology has been playing an increasingly important role in various fields of social production and life. The research results have been widely used in meteorological analysis, traffic video surveillance, autonomous driving systems.Medical image segmentation and robot vision system and other fields.In recent years, with the development of optical flow calculation, new theories and techniques for optical flow calculation models and optimization methods have been emerging, and the image sequence optical flow computing technology has made a breakthrough in computing accuracy and efficiency.However, the accuracy and robustness of optical flow computation need to be further studied when the image sequences contain difficult scenes and motion forms such as motion edge blur, non-rigid motion, brightness mutation and discontinuity.In this paper, the motion blur of image sequence is studied deeply, and the optimization model of image sequence optical flow based on guided filter is designed to overcome the complex edge structure.In order to improve the accuracy and robustness of optical flow calculation, large displacement and non-rigid motion caused by optical flow calculation edge blur and other problems.The main work of this paper includes the following points: 1.The research background and present situation of image sequence optical flow computing technology are introduced in detail, and the main problems and challenges faced by optical flow computing technology are analyzed in detail.The classical optical flow calculation methods such as HSN CLG-TV and the representative optical flow calculation models such as Classic NLNNF in recent years are introduced, and the advantages and disadvantages of various calculation methods and their application range are analyzed.Aiming at the problems of edge blur caused by complex edge structure, large displacement motion and non-rigid motion, an optical flow optimization method based on guided filter is proposed in the framework of image pyramid thinning.In the process of optical flow calculation, by guiding the linearization of filter, the details of image edge can be preserved well, and the problems such as edge blur can be overcome, so that the image edge can be enhanced and sharpened.The calculation accuracy and robustness of optical flow are improved.The typical variational optical flow methods such as HSN CLG-TV Classic NL and NNF are optimized by using standard test image sets provided by the Middlebury MPI Sintel and KITTI databases, respectively.The experimental results show that the optical flow calculation optimization model based on guided filter can effectively improve the accuracy and robustness of optical flow calculation, especially when the image sequence contains complex edge structure and large displacement motion.It can greatly improve the effect of optical flow calculation.
【學(xué)位授予單位】：南昌航空大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2017
【分類號】：TP391.41

【參考文獻】

相關(guān)期刊論文 前2條

1 張聰炫;陳震;黎明;;圖像序列變分光流計算技術(shù)研究進展[J];電子測量與儀器學(xué)報;2015年06期

2 ;Motion texture using symmetric property and graphcut algorithm[J];Journal of Zhejiang University Science A(Science in Engineering);2006年07期

，

本文編號：1719463