發(fā)布時間：2022-01-01 15:27

　　車輛軌跡數(shù)據(jù)中包含豐富的交通運行和車輛行駛特性,對于交通流理論分析和建模起到了關(guān)鍵支撐。近年來,無人機航拍成為一種高效、便捷、經(jīng)濟的交通視頻采集手段。本研究旨在構(gòu)建從無人機航拍視頻中進行高精確度車輛行駛軌跡提取的圖像處理方法,主要包括（1）采用Canny邊緣檢測法對車輛進行檢測識別;（2）基于相關(guān)內(nèi)核濾波器（KCF）的車輛跟蹤與數(shù)據(jù)校驗;（3）提出了一種基于絕對變換差和（SATD）的車輛檢測過程圖像穩(wěn)定方法。結(jié)果表明,本研究提出的方法可使檢測正確率達到451（72.6%）,相比現(xiàn)有算法具有明細提升。此外,在獲取的數(shù)據(jù)集中有368條（69.6%）軌跡存在較為顯著的檢測與跟蹤誤差。為了提高數(shù)據(jù)的可靠性以進一步完善數(shù)據(jù)庫的建立,本研究對軌跡數(shù)據(jù)進行了部分手動跟蹤,最終總共獲取了621條車輛的完整軌跡數(shù)據(jù)。 

【文章來源】：東南大學(xué)江蘇省 211工程院校 985工程院校 教育部直屬院校

【文章頁數(shù)】：108 頁

【學(xué)位級別】：碩士

【文章目錄】：
摘要
abstract
ACKNOWLEDGEMENTS
Chapter1:INTRODUCTION
    1.1.Background
    1.2.Literature review
        1.2.1.General trajectory extraction methods
        1.2.2.Applications for image processing
        1.2.3.Use of aerial videos in Transportation Engineering field and trajectory extraction
        1.2.4.Studies based on vehicle trajectory databases
    1.3.Purpose
    1.4..Research contents
Chapter2:VIDEO RECORDING AND METHODOLOGICAL FRAMEWORK
    2.1.Description of the location
    2.2.Materials and tools applied
    2.3.Video recording
    2.4.Methodological framework
Chapter3:CANNY EDGE DETECTOR APPLIED TO VEHICLE DETECTION
    3.1.Functioning of the Canny edge detector
    3.2.Vehicle detection procedure
    3.3.Limitations found during the vehicle detection process
        3.3.1.Issues regarding the color of the vehicles
        3.3.2.Issues regarding the light variation
        3.3.3.Camera displacement and the practical solution
        3.3.4.Problem on detecting lane-changers
    3.4.Vehicle detection practical process
    3.5.Noise occurrence in the vehicle detection process
        3.5.1.Failed detection
        3.5.2.Tilted trackers
        3.5.3.Trackers when only half of the vehicle was detected
    3.6.Results of the vehicle detection process
Chapter4:BLOCK MATCHING PROPOSED FOR THE CAMERA MOTION ISSUE
    4.1.Functioning of Sum of Absolute Transformed Differences(SATD)method
    4.2.A practical experiment results and limitations
        4.2.1.Displacements measured from the Area
        4.2.2.Displacements measured from the Area
    4.3.Applying the method to the case study
    4.4.Results and improvements compared to the existing vehicle detection method
Chapter5:KERNELIZED CORRELATION FILTER APPLIED TO VEHICLE TRACKING
    5.1.Functioning of the object tracker algorithm
    5.2.Difference between its proposed scope and the purpose of the present work
    5.3.Limitations of the object tracker algorithm
        5.3.0.Occlusion
        5.3.1.Background clutter
        5.3.2.Small displacements
        5.3.3.Light variation
    5.4.Vehicle tracking practical process
    5.5.Types of noise in the vehicle tracking process
        5.5.1.The tracker that loses the vehicle it was tracking
        5.5.2.Instability of the vehicle tracker
    5.6.Results of the vehicle tracking process
Chapter6:DATA CLEANSING,SMOOTHING AND VALIDATION
    6.1.Visual checking and data cleansing
    6.2.Raw data analysis and general information
    6.3.Smoothing of the traveled distances
    6.4.Smoothing results
    6.5.Data validation
        6.5.1.Macro analysis of the extracted data
        6.5.2.Deeper analysis of the extracted data
Chapter7:DATABASE CONSTRUCTION
    7.1.Database contents
    7.2.Conventional variables
    7.3.Data that was directly obtained from the video
    7.4.Variables that demanded calculation to be obtained
        7.4.1.Partial and accumulated distances
        7.4.2.Velocity
        7.4.3.Acceleration
    7.5.The final database
Chapter8:CONCLUSION
    8.1.Data extraction performance
    8.2.Achievements
    8.3.Future researches
    8.4.Database publication and use
REFERENCES
Appendix A:DATABASE SUPPORT TABLE
Appendix B:DATABASE SAMPLE



本文編號：3562409