航空影像无控拼接及匀光匀色方法研究

发布时间：2018-04-17 00:02

本文选题：航空影像 + 全局一致　；参考：《武汉大学》2017年硕士论文

【摘要】：近年来,基于星载和机载平台的遥感成像技术在地面数据快速获取、国土资源监测等应用领域有着重要地位。而实际作业过程中,研究区域涉及的范围往往不是一幅影像所能覆盖的,需要将多幅影像进行拼接形成具有较宽视角的高分辨率合成影像,以满足全局分析和整体解译等应用的需求。航空影像拼接技术流程主要包括几何对齐、色彩一致性处理,拼接线选取和边界融合四个环节。本文针对低空无人机影像序列快速无控拼接方法中的具有挑战性的几何对齐和颜色一致性校正两个算法模块进行研究。全局一致对齐。为了解决低空摄影条件下地面伪平面性带来的影像拼接累积透视变形问题,本文提出了一种能在抑制全局透视变形的同时保证局部拼接精度的通用拼接框架。首先,利用影像序列的时序关系进行尝试拼接,同时利用逐步恢复的相对几何位置探测并验证潜在的影像重叠关系,以获取更为完整的影像邻接拓扑网。特别地,无序影像数据可以预先通过从快速构建的影像相似表搜索出连通所有影像的主链转化为有序序列。然后,基于估计的影像邻接拓扑网,利用多源最短路径算法搜索出具有最少误差传递次数的参考影像,并将所有影像组织成一棵分级生成树。最后,以拓扑分析下分组估计的仿射变换模型作为初始模型参数,在基于单应模型的全局优化能量方程中加入抗透视畸变的约束项,以保持局部对齐精度与全局一致性之间的最佳平衡。颜色一致性校正。本文提出了一种有效的颜色校正方法,设计的能量函数能在最小化影像间颜色差异的同时兼顾单张影像的梯度保护和对比度优化。该方法首先利用变化检测算法识别内容变化区域以消除它对影像间对应颜色提取的干扰。然后,通过二次样条曲线对灰度映射关系的直接建模,将颜色一致性、动态范围以及梯度保护等质量约束有效地整合到统一的能量框架下。最后经凸二次规划快速地求解全局优化函数,并能保证解的全局最优性。对于全局一致拼接算法的测试,本文选取了两组覆盖范围较广的大规模低空无人机影像影像作为测试数据。实验结果表明,本文提出的算法能在保证影像序列拼接结果的全局一致性的同时,获得明显优于商业软件PTGui的局部拼接精度。此外,为了更加全面地测试本文颜色一致性处理算法的有效性,本文分选取了多时相卫星影像和航空影像的遥感数据集作为测试数据。通过与同类算法的对比,实验结果从定性和定量两方面证明了本文算法在影像拼接中对颜色一致性校正问题的有效性。
[Abstract]:In recent years, the remote sensing imaging technology based on spaceborne and airborne platform plays an important role in the field of rapid acquisition of ground data and monitoring of land resources.However, in the process of practical operation, the range of research area is often not covered by one image, so it is necessary to splice multiple images to form high-resolution composite images with wide angle of view.To meet the needs of global analysis and global interpretation and other applications.The technical flow of aviation image stitching mainly includes four links: geometric alignment, color consistency processing, stitching line selection and boundary fusion.In this paper, the challenging algorithms of geometric alignment and color consistency correction for image sequences of low altitude UAV are studied.Global consistent alignment.In order to solve the problem of image stitching cumulative perspective deformation caused by the pseudo-planarity of the ground under low altitude photography, this paper presents a universal mosaic framework which can suppress the global perspective deformation while ensuring the local stitching accuracy.Firstly, the sequential relation of image sequence is used to try to splice, and the potential image overlap relationship is detected and verified by using the gradually restored relative geometric position to obtain a more complete image adjacent topology network.In particular, the unordered image data can be pre-searched from the rapidly constructed image similarity table to transform the main chain of all the images into an ordered sequence.Then, based on the estimated image adjacent topology network, the multi-source shortest path algorithm is used to search the reference images with minimum error transfer times, and all the images are organized into a hierarchical spanning tree.Finally, taking the affine transformation model of grouping estimation under topological analysis as the initial model parameter, the global optimization energy equation based on the monoclinic model is added to the global optimization energy equation with constraints against perspective distortion.In order to maintain the best balance between local alignment accuracy and global consistency.Color consistency correction.In this paper, an effective color correction method is proposed. The designed energy function can minimize the color difference between images while taking into account the gradient protection and contrast optimization of single image.Firstly, the change detection algorithm is used to identify the region of content change to eliminate the interference to the corresponding color extraction between images.Then, through the direct modeling of the gray mapping relation by the quadratic spline curve, the quality constraints such as color consistency, dynamic range and gradient protection are effectively integrated into the unified energy framework.Finally, the global optimization function is solved quickly by convex quadratic programming, and the global optimality of the solution is guaranteed.In this paper, two sets of large scale low altitude UAV images covering a wide range are selected as test data for the test of global consistent stitching algorithm.The experimental results show that the proposed algorithm can ensure the global consistency of the image sequence stitching results and obtain the local stitching accuracy which is obviously superior to the commercial software PTGui.In addition, in order to test the validity of the algorithm, the remote sensing data sets of multitemporal satellite images and aerial images are selected as the test data.Compared with the similar algorithms, the experimental results demonstrate the effectiveness of the proposed algorithm in color consistency correction in image stitching from both qualitative and quantitative aspects.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP751

【参考文献】