隨著公共場所和與工作相關(guān)的場所中安全攝像機的日益普及,對視覺對象跟蹤的需求也越來越大。目前,目標(biāo)跟蹤正在被很多行業(yè)應(yīng)用,例如視頻監(jiān)控、交通流量監(jiān)控、智能環(huán)境等。在目前的計算機視覺相關(guān)研究中,目標(biāo)跟蹤是圖像領(lǐng)域中最具挑戰(zhàn)性和新興的主題之一。在目標(biāo)跟蹤中,有許多影響跟蹤性能的因素,包括目標(biāo)間遮擋、目標(biāo)與背景間遮擋、照明變化、動態(tài)背景以及幀噪聲等。本文結(jié)合了室內(nèi)和室外環(huán)境中的幾種方法來解決上述問題。首先,本文分析了具有魯棒性和簡單設(shè)計特性的均值漂移跟蹤(Mean-Shift Tracking,MST)算法。但是,MST在具有挑戰(zhàn)的條件(例如遮擋和照明參數(shù)變化)中使用時會有一些限制。為了解決部分和完全遮擋問題,本文提出了基于圖像相似性度量(Image Similarity Measures,ISM)的單目標(biāo)跟蹤(Single Object Tracking,SOT)方法,包括歸一化互相關(guān)(Normalized Cross-Correlation,NCC)和NCC擴展方法。在不同的室內(nèi)和室外場景下的實驗結(jié)果證明了本文針對各種跟蹤參數(shù)采用權(quán)衡策略后所提出方法的有效性。因此,本文在遮擋相關(guān)場景中實現(xiàn)...

【文章頁數(shù)】：85 頁

【學(xué)位級別】：碩士

【文章目錄】：
摘要
Abstract
Chapter1 Introduction
    1.1 Purpose and Significance of the Research
    1.2 Background of the Research
    1.3 Breifly Analysis of Research Status at Abroad and Foreign
        1.3.1 Research Status at Abroad
        1.3.2 Research Status at Home
        1.3.3 A Brief Analysis and Literature Review of Domestic and Foreign
    1.4 Objectives and Thesis Outline
        1.4.1 Objectives
        1.4.2 Thesis Outline
Chapter2 Motion Detection and Mean Shift
    2.1 Introduction
    2.2 Background Subtraction
        2.2.1 Background Subtraction for Each Color
        2.2.2 Masking Technique
        2.2.3 Tracking Through Edge Detection and Boundary Delineation
    2.3 Mean Shift Tracking
        2.3.1 Intuitive Idea of Mean Shift Tracking
        2.3.2 Tracking Mean Shift Object Tracking
        2.3.3 MST Implementation Procdure
    2.4 Summary
Chapter3 Image Similarity Measures
    3.1 Introduction
    3.2 Tracking using Image Similarity Measures
        3.2.1 ISM-based MSD Method
        3.2.2 ISM-based Histogram Matching
    3.3 Occlusion Handling based on ISM
        3.3.1 ISM-based NCC Method
        3.3.2 ISM-based Extend NCC Method
    3.4 Performance of ISM
    3.5 Summary
Chapter4 Multi-object Tracking Method
    4.1 Introduction
    4.2 Object Representation and Gaussian Mixture Model
        4.2.1 Object Representation
        4.2.2 Gaussian Mixture Model
        4.2.3 Results and Analysis of GMM
    4.3 Proposed Method for MOT in Outdoor Environment
        4.3.1 Object Segmentation Process
        4.3.2 Contours
        4.3.3 Adaptive Background Subtraction with Running Average
        4.3.4 Image Morphology
    4.4 MOT Results and Discussion
    4.5 Summary
Conclusions
References
Acknowledgement



本文編號：3738000