【正文】 西北工業(yè)大學(xué)明德學(xué)院 本科畢業(yè)設(shè)計(jì)論文 I 目 錄 摘 要 .............................................................................................................................. 1 ABSTRACT................................................................................................................... 2 第一章 緒 論 ................................................................................................................ 4 課題研究背景和意義 ....................................................................................... 4 國內(nèi)外研究現(xiàn)狀 ............................................................................................... 5 本文的具體結(jié)構(gòu)安排 ....................................................................................... 7 第二章 運(yùn)動(dòng)目標(biāo)檢測(cè) .................................................................................................. 8 檢測(cè)算法及概述 ............................................................................................... 8 連續(xù)幀間差分法 ..................................................................................... 9 背景去除法 ........................................................................................... 11 光流法 ................................................................................................... 13 第三章 運(yùn)動(dòng)目標(biāo)跟蹤方法 ........................................................................................ 16 引言 ................................................................................................................. 16 運(yùn)動(dòng)目標(biāo)跟蹤方法 ......................................................................................... 16 基于特征匹配的跟蹤方法 ................................................................... 16 基于區(qū)域匹配的跟蹤方法 ................................................................... 17 基于模型匹配的跟蹤方法 ................................................................... 18 運(yùn)動(dòng)目標(biāo)搜索算法 ......................................................................................... 18 絕對(duì)平衡搜索法 .................................................................................... 18 絕對(duì)平衡搜索法實(shí)驗(yàn)結(jié)果 ............................................................................. 19 歸一化互相關(guān)搜索法 ........................................................................... 21 歸一化互相關(guān)搜索法實(shí)驗(yàn)結(jié)果 及 分析 ......................................................... 22 第四章 模板更新與軌跡預(yù)測(cè) .................................................................................. 26 模板更新簡(jiǎn)述及策略 ..................................................................................... 26 軌跡預(yù)測(cè) ......................................................................................................... 28 線性預(yù)測(cè) ............................................................................................... 29 平方預(yù)測(cè)器 ........................................................................................... 30 西北工業(yè)大學(xué) 本科畢業(yè)設(shè)計(jì)論文 II 實(shí)驗(yàn)結(jié)果及分析： ......................................................................................... 31 致 謝 ............................................................................................................................ 36 參考文獻(xiàn) ...................................................................................................................... 37 畢業(yè)設(shè)計(jì)小結(jié) .............................................................................................................. 38 西北工業(yè)大學(xué)明德學(xué)院 本科畢業(yè)設(shè)計(jì)論文 1 摘 要 圖像序列目標(biāo)跟蹤是計(jì)算機(jī)視覺中的經(jīng)典問題，它是指在一組圖像序列中，根據(jù)所需目標(biāo)模型，實(shí)時(shí)確定圖像中目標(biāo)所在位置的過程。它最初吸引了軍 方的關(guān)注，逐漸被應(yīng)用于電視制導(dǎo)炸彈、火控系統(tǒng)等軍用備中。序列圖像運(yùn)動(dòng)目標(biāo)跟蹤是通過對(duì)傳感器拍攝到的圖像序列進(jìn)行分析，計(jì)算出目標(biāo)在每幀圖像上的位置。它是計(jì)算機(jī)視覺系統(tǒng)的核心，是一項(xiàng)融合了圖像處理、模式識(shí)別、人工只能和自動(dòng)控制等領(lǐng)域先進(jìn)成果的高技術(shù)課題，在航天、監(jiān)控、生物醫(yī)學(xué)和機(jī)器人技術(shù)等多種領(lǐng)域都有廣泛應(yīng)用。因此，非常有必要研究運(yùn)動(dòng)目標(biāo)的跟蹤。 本論文就圖像的單目標(biāo)跟蹤問題，本文重點(diǎn)研究了幀間差分法和背景去除法等目標(biāo)檢測(cè)方法，研究了模板相關(guān)匹配跟蹤算法主要是：最小均方誤差函數(shù)(MES)，最小平均絕對(duì)差 值函數(shù) (MAD)和最大匹配像素統(tǒng)計(jì) (MPC)的跟蹤算法。在跟蹤過程中， 由于跟蹤設(shè)備與目標(biāo)的相對(duì)運(yùn)動(dòng) , 視野中的目標(biāo)可能出現(xiàn)大小、形狀、姿態(tài)等變化 , 加上外界環(huán)境中的各種干擾 , 所要跟蹤的目標(biāo)和目標(biāo)所在的場(chǎng)景都發(fā)生了變化 , 有可能丟失跟蹤目標(biāo)。為了保證跟蹤的穩(wěn)定性和正確性 , 需要對(duì)模板圖像進(jìn)行自適應(yīng)更新。由于目標(biāo)運(yùn)動(dòng)有一定得規(guī)律，可以采取軌跡預(yù)測(cè)以提高跟蹤精度，本文采用了線性預(yù)測(cè)法。 對(duì)比分析了相關(guān)匹配算法的跟蹤精度和跟蹤速度；對(duì)比不采用模板更新和模板跟新的跟蹤進(jìn)度和差別，實(shí)驗(yàn)表明，跟蹤算法加上軌跡 預(yù)測(cè)及模板跟新在很大程度上提高了跟蹤幀數(shù)，提高了跟蹤精度，具有一定的抗噪聲性能。 關(guān)鍵詞 ： 目標(biāo)跟蹤，目標(biāo)檢測(cè) ，軌跡預(yù)測(cè) ，模板更新 西北工業(yè)大學(xué) 本科畢業(yè)設(shè)計(jì)論文 2 ABSTRACT Target tracking, image sequence is a classic puter vision problems, it is defined as a set of image sequences, in accordance with requirements of the target model, realtime images to determine the location of the target process. It initially attracted the concern of the military has gradually been applied to televisionguided bombs, fire control systems for military preparation. Moving target tracking sensor is taken through the image sequence analysis, to calculate the target image in each frame position. It is the core of puter vision system is a bination of image processing, pattern recognition, artificial only and the results of automatic control in areas such as advanced hightech issues in the aerospace, control, biomedical and robotics fields, etc. There are widely used. Thus, it is necessary to study the tracking of moving targets. In this paper, the image of the singletarget tracking problem, research the target detection method is mainly based on interframe difference and background removal method to detect the target in preparation for target tracking. Template matching tracking algorithm is: the smallest mean square error function (MES), the smallest mean absolute difference function (MAD) and the maximum matching pixel statistics (MPC) of the tracking algorithm. In the tracking process, due to the relative camera movement with the goal, the goal of vision may occur in size, shape, gesture, such as changes in the external environment bined with the various kinds of interference, as well as over time, to track where the goals and objectives scene changes have taken place, it is possible to track the target is lost. In order to ensure the stability and tracking accuracy, the need for adaptive