【作者】 张宁；

【导师】 王竞雪；

【作者基本信息】 辽宁工程技术大学 ， 测绘科学与技术， 2021， 硕士

【摘要】 随着社会的不断发展进步与日益增大的市场需要,三维重建已经在社会的多个方面展现出重要的角色,传统的三维重建方案面对人们日益严苛的要求已经难以为继,基于此基于影像的三维重建方案由于门槛低、数据多样和成本低等优势,被大多数学者所喜爱。然而现有的三维重建中所存在的匹配点数目稀疏、分布不均、未考虑直线特征区域和精度不足等问题,本文提出了一种基于三角约束和传播的分层点准稠密匹配算法,该算法将稀疏匹配点构建三角网,并提取参考影像直线特征,以三角网与直线段交点和三角网中点作为匹配基元,构建子区域重叠描述符,联合欧氏距离作为判定依据,面对匹配丢失点,则会有多约束条件进行二次匹配。根据实验结果,本文获得可靠的密集匹配点,并结合三维重建,可以获得良好的视觉体验。本文的主要研究内容如下:(1)子区域重叠描述符构建:以目标点所在像素索引为中心,建立描述区域,通过使用目标点所在像素索引和四个角点点坐标来设置子区域的对角线坐标,相邻的子区域间会存在部分重叠,紧接着将四个子区域再次分成为四个块区域,联合高斯加权统计四个块区域的梯度直方图,然后进行均值向量和标准差向量的归一化,最后将四个子区域的描述符向量连接,得到子区域重叠描述符向量,联合欧式距离进行匹配点的判定。(2)多约束匹配:若初次匹配失败,则进行多约束条件下的匹配,多约束条件匹配是对描述符约束、极线约束、仿射变换和三角形约束分别赋予相应权重,构建得分计算公式,其中得分最高且超过设定阈值者为匹配成功点。(3)实验验证与分析:分别进行了多组不同影像准稠密匹配算法实验,不同算法的准稠密匹配算法实验,详细叙述参数选择,通过对实验结果的对比分析,验证了本文提出的准稠密匹配算法的良好适用性与稳定性,同时进行了影像稀疏重建和结合本文算法的稠密重建的实验,从侧面二次证明了本文准稠密匹配算法的鲁棒性与普适性。该论文有图56幅,表6个,参考文献112篇。更多还原

【Abstract】 With the continuous development and progress of society and the increasing market demand,3D reconstruction has played an important role in many aspects of society.The traditional 3D reconstruction scheme has been difficult to sustain in the face of increasingly stringent requirements.Because of the advantages of low threshold,data diversity and low cost,the image-based 3D reconstruction scheme is loved by most scholars.However,there are some problems in the existing 3D reconstruction,such as sparse number of matching points,uneven distribution,not considering the line feature area and insufficient accuracy.In this paper,a hierarchical point quasi dense matching algorithm based on triangular constraint and propagation is proposed.The algorithm constructs a triangular network from sparse matching points,and extracts the line features of the reference image,and uses the intersection of the triangular network and the line segment and the midpoint of the triangular network As matching primitives,the overlapping descriptors of sub regions are constructed,and the Euclidean distance is used as the judgment basis.According to the experimental results,this paper obtains reliable dense matching points,and combined with 3D reconstruction,can obtain good visual experience.The main contents of this paper are as follows:(1)Sub region overlap descriptor Construction: take the pixel index of the current point as the center,establish the description region,set the diagonal coordinates of the sub region by using the pixel index of the target point and the coordinates of the four corner points,there will be partial overlap between the adjacent sub regions,then divide the four sub regions into four block regions,and count the four block regions with Gaussian weighting Then the mean vector and standard deviation vector are normalized.Finally,the descriptor vectors of four sub regions are connected to get the overlapping descriptor vectors of sub regions,and the matching points are determined by combining Euclidean distance.(2)Multi constraint matching: if the initial matching fails,the matching under multi constraint conditions will be carried out.Multi constraint matching is to give corresponding weights to descriptor constraint,epipolar constraint,affine transformation and triangle constraint respectively,and construct score calculation formula.The one with the highest score and exceeding the set threshold is the success point of matching.(3)Experimental verification and analysis: several groups of different image quasi dense matching algorithm experiments and quasi dense matching algorithm experiments of different algorithms are carried out respectively,and the parameter selection is described in detail.Through the comparative analysis of the experimental results,the good applicability and stability of the quasi dense matching algorithm proposed in this paper are verified.At the same time,image sparse reconstruction and dense re matching combined with this algorithm are carried out The experimental results show the robustness and universality of the quasi dense matching algorithm.This paper has 56 pictures,6 tables and 112 references.更多还原