复杂地形下耦合多基元的低空倾斜立体影像匹配研究

Oblique low-altitude photogrammetry has become a hot topic and a popular field of research. Compared with vertical photography, low-altitude photography is conducive to improve the intersection accuracy by wide-baseline and large tilt angle photogrammetry. In the previous research, the surface features problems of overlay imaging, geometric large distortion and scale change etc were found between oblique images with complex terrain, and most of image matching algorithms may be difficult to obtain high-precision and evenly distributed corresponding points, so automatic image matching should be further studied. The project on the basis of previous research, starting from oblique imaging mechanism, focusing on the theory and method of image matching with complex terrain, further and systematically researching on multi-primitive and local perspective invariant descriptor based image matching method and strategy. The main contents as follows: theoretical analysis and simulation of image geometric distortion with complex terrain, exploring the rules of feature location deviation and matching measure window shape change; research on high-precision Harris-Laplace corner location error compensation mechanism, perspective transformation model of matching adaptive measure window shape by TIN, combing triangle network constrains, epipolar geometry and least-square method, and a point-polygon multi-primitive coupled coarse-to-fine oblique low-altitude image matching with complex terrain is proposed and verified. The research results will enrich and develop the theory and method of oblique low-altitude image matching with complex terrain, and lay the theories and technical foundations for image intelligent processing of irregular photogrammetry.

低空倾斜摄影测量已成为研究的热点，相比垂直摄影方式，能实现宽基线大倾角的摄影测量，有利于提高交会精度。申请者在前期研究中发现，复杂地形下倾斜影像之间存在地物叠置成像、几何变形和尺度变化大等问题，大多数匹配算法难于获得高精度分布均匀的同名点，影像自动匹配研究有待于进一步深入。本课题拟从倾斜成像方式出发，围绕复杂地形下影像匹配的理论与方法，深入系统地研究基于多特征基元与局部透视不变描述子的匹配模型与策略。具体内容包括：理论分析与仿真复杂地形下影像几何变形，探索特征定位偏差与匹配测度窗口形状变化规律；研究高精度Harris-Laplace点位偏差补偿机制和TIN格网构建匹配测度窗口自适应透视变换模型等，结合三角网约束、核线与最小二乘法创建耦合点面特征多基元由粗到精的影像匹配方法。研究成果将丰富并完善复杂地形下影像匹配的理论与方法，为低空倾斜摄影测量等不规则测量方式影像智能处理奠定理论与技术基础。

在低空倾斜摄影测量中，平台飞行高度低地形起伏相对较大，其引起的影像几何变形会严重影响匹配性能，采用单一的仿射变化关系可能难于表述立体像对间各个局部的扭曲变化，已不适用于地表建筑物高低错落、山区等复杂地形下低空倾斜摄影测量影像匹配。本项目围绕复杂地形下影像匹配存在的问题，构建了由粗到精的高效匹配策略、影像几何变形与匹配测度窗口形状之间的相关性、点位偏差补偿模型、局部透视不变性特征检测模型。通过对低空倾斜影像匹配的理论与方法进行研究，实现了适合包括平地、山区等各种复杂地形的低空倾斜立体影像自动匹配，并设计出了耦合点面特征多基元的影像匹配算法与技术方案。从区域边界与重心定位方面改进了MSER算法进行影像粗匹配，研究构建三维表面模型和严密的相对定向模型，建立了一种基于优化MSER面特征基元的倾斜立体影像粗匹配方法，从试验对比分析表明：优化后MSER得到的匹配点解算视差中误差比未优化的MSER相对定向小得多；利用Harris-Laplace算法，考虑Harris角点响应函数与面特征获得透视变换几何变形特征，构建误差方程，通过迭代精化，获得最佳的亚像素Harris-Laplace角点，通过复杂地形下倾斜影像试验结果表明：该算法对于旋转、尺度变化较大的影像可获得匹配点数量可观且分布均匀，匹配结果非常理想。传统基于灰度的影像匹配方法和SIFT算法在试验数据里均失败，而ASIFT算法只获得少于本课题2/3数量的匹配点却耗时超过3倍。试验结果表明，本项目研究方法可实现宽基线大倾角的低空倾斜影像匹配，在复杂地形低空影像匹配中具有较强的实用性，可在一定程度上提高低空影像匹配的成功率和可靠性。研究成果对解决高精度低空倾斜摄影测量数据自动化处理具有积极的意义和重要的实用价值。