基于无人机航拍图像的树冠三维重建3DReconstructionofTreeCrownBasedonUnmannedAerialVehicleAerialImage学科专业:计算机科学与技术研究生:陆泽萍指导教师:张冰怡副教授天津大学计算机科学与技术学院二零一三年十一月独创性声明本人声明所呈交的学位论文是本人在导师指导下进行的研究工作和取得的研究成果,除了文中特别加以标注和致谢之处外,论文中不包含其他人已经发表或撰写过的研究成果,也不包含为获得天津大学或其他教育机构的学位或证书而使用过的材料。与我一同工作的同志对本研究所做的任何贡献均已在论文中作了明确的说明并表示了谢意。学位论文作者签名:签字日期:年月日学位论文版权使用授权书本学位论文作者完全了解天津大学有关保留、使用学位论文的规定。特授权天津大学可以将学位论文的全部或部分内容编入有关数据库进行检索,并采用影印、缩印或扫描等复制手段保存、汇编以供查阅和借阅。同意学校向国家有关部门或机构送交论文的复印件和磁盘。(保密的学位论文在解密后适用本授权说明)学位论文作者签名:导师签名:签字日期:年月日签字日期:年月日摘要随着无人机技术的发展,无人机已成为一个重要的航空器,无人机航拍也成为空间数据源获取的重要途径,在军事和民用领域中都成为不可替代的测绘手段。但是无人机航拍的序列图像仅仅包含二维信息,无法再现实体的三维几何结构,缺乏直观性。所以,为了获得更准确、更详细的信息,无人机航拍在应用上不可避免的要从二维走向三维。而树类物体由于本身结构的复杂性,使得对其的三维重建更具有研究意义。本文针对无人机在山西某山区航拍获得的图像,提出了一个完整的树冠三维重建算法。无人机航拍图像由于航拍高度的影响,所包含的树冠纹理、轮廓信息贫乏,针对这个问题,本文提出了基于分水岭分割的特征区域提取方法来充分提取出能反映树冠结构的特征点,并计算区域相关系数对特征区域进行准确地匹配。此外,无人机航拍过程中飞机的抖动、倾斜等造成摄像机的倾斜、抖动等,使得有航拍摄像机的成像模型具有本身的独特性,本文在三维重建过程中,针对其独特性确定模型内外参数,并利用双目立体视觉原理计算特征点集的深度信息;在目标建模上,树类物体本身结构复杂,其丰富的枝干信息决定了树木的结构,因此,本文采用了模拟树木生长方式的L系统规则进行树冠的建模。本文对从无人机航拍图像手动截取的树冠图像进行实验,证明本文的方法能够充分提取树冠结构特征点并匹配特征点,从航拍图像的有限信息中重建出树冠近似三维模型。提出的特征点提取和匹配方法充分考虑到航拍图像和待重建目标的特点,对今后的相关研究提供了一定的参考价值;在建模方法上,考虑到了树冠的结构特点,充分利用了已有的信息,在更高难度的问题上达到了较好的效果,是一种很好的树冠建模思路。关键词:三维重建航拍图像分水岭分割区域相关系数L系统ABSTRACTAsanimportantpartoftheaircraft,unmannedaerialvehicle(UAV)aerialhasbecomeanimportantmethodtoobtainspacedata.Inthefieldofmilitaryandcivilian,UAVbecomesanirreplaceablemeansofsurveyingandmapping.TheUAVaerial2Dimageswithout3Dsolidgeometrydataoftheobject,lackofintuitive,soinordertoobtainmoreaccurateanddetailedinformation,UAVvisualinevitablyfrom2Dto3Dinapplications.3DReconstructionoftreemakesmoreresearchsignificanceduetoitscomplexstructure.Inthispaper,weputforwardacompletecrown3DreconstructionalgorithmbasedontheUAVaerialimagestakenfromacertainmountainareainShanxi.Consideringtheaerialimagecontainslimitedcanopytextureandcontourinformation,weproposedafeatureextractmethodandmatchmethodbasedonLocalAreaCorrelationCoefficient(LACC),inordertofullextractthefeaturepointsthatcanreflectthestructureofthetargetandmatchthefeaturepointsaccurately.Furthermore,aerialimagingmodelhassomepeculiarityduetothetiltingandjittercausedbyshakingandtiltsofUAV,weestimatedinsideandoutsideparameterstargetedtothismodelandcalculatethe3Dinformationofthefeaturepointsbasedontheprincipleofbinocularstereovision.Treehasitsuniqueshapecharacteristicsdeterminedbyitsownbranchesstructure,soweproposedatreemodelingmethodbasedonLsystemtoSimulatetreegrowth.DotheexperimentsonthecanopyimagecutfromtheUAVaerialimagemanually,wefoundthatthismethodcanfullyextractthefeaturepointsofcanopystructureandmatchingfeaturepointsaccurately,andfinallyreconstructtheapproximate3Dmodelofthetreefromthelimitedinformationofaerialimages.Thefeaturepointsextractionandmatchingmethodinthispaperfullyconsideredthecharacteristicsofaerialimagesandthetargettoreconstruction,providedsomereferencesignificanceforthefutureworks;whilethemodelingmethod,consideringthecanopystructurecharacteristics,makefulluseoftheexistinginformation,onahigherdifficultyproblemachievesagoodeffect,isagoodtreemodelingideas.KEYWORDS:3DReconstruction,aerialimage,WatershedSegmentation,LACC,LSystem目录第一章绪论................................................................................................................11.1课题背景..........................................................................................................11.2研究内容..........................................................................................................21.3全文安排..........................................................................................................3第二章序列图像的树冠三维重建研究现状...........................................................42.1树木三维重建研究现状...................................................................................42.2双目立体视觉...................................................................................................42.3特征点提取和匹配的研究现状.......................................................................52.4摄像机标定的研究现状...................................................................................62.5三维建模相关工作...........................................................................................72.6本章小结...........................................................................................................7第三章特征点提取和匹配........................................................................................93.1图像获取和预处理..........................................................................................93.1.1无人机航拍图像....................................................................................93.1.2图像预处理..........................................................................................103.2特征点提取和匹配方法.................................................................................113.2.1基于分水岭分割的特征提取方法......................................................123.2.2基于RGB颜色空间中的区域相关系数的区域匹配.........................143.2.3基于最相关和次相关比例法去除误匹配...........................................163.2.4特征点匹配的准确性评估.................................