星空背景下的目标检测与跟踪

国防科学技术大学硕士学位论文星空背景下的目标检测与跟踪姓名：张路平申请学位级别：硕士专业：信息与通信工程指导教师：李飚2010-11国防科学技术大学研究生院硕士学位论文第i页摘要利用探测器对空间目标进行光学观测，由于探测器在运行过程中暴露在外层空间辐射下，获取的星空图像除了受探测器内部噪声影响外，还受空间辐射噪声的干扰。恒星和远距离的运动目标在探测器焦平面上所成的像为分布在较暗背景上的点状光斑，无结构及纹理特征且只占有有限个像素。由于图像中存在恒星和噪声的干扰，使得有效探测弱小目标并提取其运动轨迹成为一项具有挑战性的工作，本文围绕这个课题展开研究。本文研究了EMCCD的工作原理及温度增益特性，分析了其噪声分布情况。从均值、方差、相关系数等统计量入手分析了星空图像背景噪声的时空域采样特性。利用行列均值相减法对图像进行非均匀性校正，去除固定噪声，将校正后的图像序列视为一个平稳的二维随机过程。分析了运动目标的空间散布、时间起伏特征，将目标在运动过程中的灰度变化表征成一种围绕固定值发生正弦变化的过程，以此为基础建立星空图像场景模型及目标信号模型。将星空背景下的目标与恒星的检测看成是一个二元假设检验问题，提出了以广义似然比判决为基础的昀优检测方法来提取星图中的恒星和目标。本文利用星背景抑制的策略去除恒星，将目标从星图中筛选出来，实现目标与恒星的有效分离。为了实现星背景抑制目的，本文提出两种算法来识别星图提取星点。第一种算法是基于结构信息变化的星图识别算法，该算法利用星点间拓扑结构相对稳定不变的特性，选取马氏距离与中心化的相关系数作为结构特征相似性测度来表征星点间的结构信息，通过结构信息的变化来判断是星点还是目标，从而达到星图识别的目的。第二种算法是以改进的三角形算法为基础的星图匹配算法。利用上述两种算法将星图中的星点提取出后，以星点为背景生成掩膜图像，利用背景减除法将星抑制掉，得到真正感兴趣的目标并对其进行连续跟踪。针对识别过程中出现恒星数目不定的情况，本文提出了基于结构信息和反查导航星库的识别策略，并详细研究了当目标与星靠近时利用角距信息无法将目标与星分开时的伪Hough微分识别算法。主题词：星空环境，星背景抑制，伪Hough微分识别算法，目标检测跟踪国防科学技术大学研究生院硕士学位论文第ii页ABSTRACTAsthedetectorsoperateintheouterspaceradiationexposure,thestarimagesareaffectedbyspaceradiationnoiseinterferenceexcepttheinternalnoisewhenusingthedetectorstogetopticalobservationsofspacetargets.Thefixedstarandmovingtargetsfromfarawayimageasthedistributionofdarkspotsonthespotofthedetectorsfocalplane,withonlyafinitenumberofpixelsbutwithoutstructureandtextureinformation,Undertheinfluenceofnoiseandtheinterferencetheofthevastfixedstar,itbecomesacomplexandchallengingtaskwhiledetectingthesmalltargetsandextractingtheirtrajectoryseffectively.Thisproblemisstudiedinthispaperaroundthistopic.Inthispaper,theworkingprincipleandtherelationbetweentemperatureandgaincharacteristicsoftheEMCCDisresearched,thenoisedistributionandthespatio-temporalamplingcharacteristicsofthestarimageisanalysedfromthemean,variance,correlationcoefficientofthestatistics.Afterusingthemeansofsubtractingthemeanranktorevisethenon-uniformityoftheimageandremovethefixednoise,therevisedimagesequencesareregardedasatwo-dimensionalstationaryrandomprocess.Thespatialdistributionandthetimefluctuationcharacteristicsofthemovingobjectisanalysed.Whenthetargetsmove,theprocessoftheirgray-scalechangingischaracterizedasoccuringaroundastationaryvalueofasinusoidalprocess.Basedonthisresult,itestablishestheStarimagescenemodelandthetargetsignalmodel,regardstheproblemofdetectingtargetandfixedstarunderstarbackgroundasabinaryhypothesistestingproblem,andproposesthebestmethodbasedongeneralizedlikelihoodratiotesttoextractthestarchartandobjectives.Thestrategyofsuppressingstarsbackgroundisgiventoremovethefixedstar,toselectthetargetfromthechart,andtoseparatethetargetfromstarseffectively.Inordertodistinguishstarsfromtargets.Twoalgorithmstoextractthestarpointsandrecognisethestarchartsareproposed.Thefirstalgorithmisbasedonthestructuralinformationofthestarpattern,accordingtothecharacteristicsthatthestartopologyisrelativelystablebetweenthesamefeatures,whichselectstheMarkovdistanceandthecenterofcorrelationcoefficientasthestructuresimilaritymeasuretocharacterizethestructuralinformationbetweenthestarpoint,usestructuralinformationtoidentifywhetherit’sthestarpointorthetarget,soastoachievethepurposeofstarpatternrecognition.Thesecondalgorithmisthepatternmatchingrecognitionalgorithmwhichbasedontheimprovedtriangularalgorithm.Afterextractingthestarpointsusingthosealgorithmsabove,itgeneratesthebackgroundmaskimagewiththeextractingstars,usesthebackgroundsubtractionmethodtosuppressoffstars,getsrealinterestedobjectivesanditscontinuoustrackings.Forthesituationthatthenumbersofstarsareuncertainduringidentifyingtargets,国防科学技术大学研究生院硕士学位论文第iii页thestrategybasedonstructuralinformationchangeandanti-recognitionguidestarcatalogisproposed.ItgivesadetailedstudyonthePseudoHoughDifferentialRecognitionAlgorithmwhenthetargetisclosetothestar,whichisdifficulttoseparatefromthemonlyusingtheangulardistanceinformation.KeyWords：starenvironment，starbackgroundsupperess，PseudoHoughDifferentialRecognitionAlgorithm，targetdetectingandtracking国防科学技术大学研究生院硕士学位论文第III页表目录表2.1非均匀性校正前后图像性能参数....................................................................16表2.2相关系数的统计值............................................................................................16表3.1校正前后图像的性能参数................................................................................27表3.2图像累计前后性能参数....................................................................................29表4.1马氏矩阵............................................................................................................37表4.2相关矩阵............................................................................................................37表4.3两个不同时刻的马氏距离与相关系数............................................................39表4.4不同运动速率下的增量变化值........................................................................44表4.5星图识别的基本性能........................................................................................51国防科学技术大学研究生院硕士学位论文第IV页图目录图2.1EMCCD增益与温度、温度间的关系...............................................................7图2.2时域采样星图及其三维分布.........