基于视频流的室内外环境多目标跟踪方法研究

发布时间：2023-02-08 16:26

　　随着公共场所和与工作相关的场所中安全摄像机的日益普及,对视觉对象跟踪的需求也越来越大。目前,目标跟踪正在被很多行业应用,例如视频监控、交通流量监控、智能环境等。在目前的计算机视觉相关研究中,目标跟踪是图像领域中最具挑战性和新兴的主题之一。在目标跟踪中,有许多影响跟踪性能的因素,包括目标间遮挡、目标与背景间遮挡、照明变化、动态背景以及帧噪声等。本文结合了室内和室外环境中的几种方法来解决上述问题。首先,本文分析了具有鲁棒性和简单设计特性的均值漂移跟踪(Mean-Shift Tracking,MST)算法。但是,MST在具有挑战的条件(例如遮挡和照明参数变化)中使用时会有一些限制。为了解决部分和完全遮挡问题,本文提出了基于图像相似性度量(Image Similarity Measures,ISM)的单目标跟踪(Single Object Tracking,SOT)方法,包括归一化互相关(Normalized Cross-Correlation,NCC)和NCC扩展方法。在不同的室内和室外场景下的实验结果证明了本文针对各种跟踪参数采用权衡策略后所提出方法的有效性。因此,本文在遮挡相关场景中实现...

【文章页数】：85 页

【学位级别】：硕士

【文章目录】：
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



本文编号：3738000