【作者】 陆宏伟；

【导师】 于起峰；

【作者基本信息】 国防科学技术大学 ， 航空宇航科学与技术， 2004， 博士

【摘要】 在现代卫星遥感对地观测中,由多颗小卫星组成的卫星编队具有比单颗卫星更优越的性能,能有效地提高各种测量的空间分辨率和时间分辨率。目前,基于卫星编队遥感图像的对地定位技术,已成为当前摄影测量与遥感领域中的研究热点之一。论文针对基于卫星编队遥感图像对地定位中的遥感卫星编队轨道设计、摄像机相对定向、摄像机姿态确定、图像特征定位和图像匹配等关键问题进行了深入研究,取得了一系列研究成果。 系统地研究了圆轨道卫星绕飞编队的相对运动分析和轨道设计方法。针对一般运动学方法的不足,提出了一种基于零倾角轨道变换的运动学新方法,并用这种运动学新方法分析了参考卫星轨道倾角对环绕卫星轨道根数的影响。根据遥感测量对卫星轨道的要求,设计了一种覆盖范围大、高度基本相等的卫星椭圆绕飞编队。研究了在现有星载摄像机外方位参数测量水平下,直接定位方法所能达到的定位精度。 星载摄像机的相对定向和姿态确定是实现对地定位的关键问题。小视场角条件下的摄像机相对定向研究表明:基于立体像对本质矩阵分解的线性算法难以得到合理的相对定向结果,而基于多视图仿射矩阵分解的线性算法可以获得稳定性较好的相对定向结果,并可用光束法平差进一步提高相对定向精度。在摄像机相对定向的基础上,提出了一种基于卫星编队基线观测向量的摄像机姿态确定算法,为实现对地直接定位提供了必要条件。研究了基于卫星动力学模型的摄像机姿态估计算法。结果表明:摄像机姿态确定精度主要取决于摄影基线观测向量,即相对定向的平移参数的精度;利用非线性预测滤波可以有效地提高摄像机姿态确定精度。 针对小基高比造成直接定位高程定位误差较大的情况,研究了利用多种相对控制进行联合平差提高定位精度的可行性,并与有少量地面控制点时联合平差的定位结果进行了比较。讨论了卫星网中的定位结果转化为地面网的方法和空间投影与测绘制图问题。 在对现有图像特征提取算法研究基础上,提出了一种基于多特征复合的节点特征快速提取算法。研究了图像噪声、特征尺度和特征属性对节点特征亚像素定位精度的影响,并分析了道路特征两侧灰度差对其亚像素定位精度的影响。 从可靠性、精度和效率等三个方面对基于区域的图像匹配算法进行了深入研究。兼顾可靠性和精度两方面的要求,提出了一种匹配特征区域的快速最优选择算法;针对影响匹配效率的两个主要因素——搜索空间和搜索策略,分别提出了基于矩不变量和奇异值分解的特征点匹配算法和基于改进遗传算法的匹配搜索策略,有效地减小了匹配搜索空间并提高了匹配效率。更多还原

【Abstract】 In modern earth observation by satellite-based remote sensing, clusters of low-performance satellite flying in formation may provide enhanced performance over single high-performance satellite. This is especially true for photogrammetry and remote sensing based on the satellite formation where stereographic imaging may provide higher spatial resolution and temporal resolution. The high accuracy ground position based on remote sensing image from satellite formation had been being a hot topic of digital photogrammetry and remote sensing. This dissertation is to study on the algorithms for ground position based on remote sensing images from satellite formation. The results attained are as follows.The relative motion analysis and orbit design of satellite formation flying on circular orbits have investigated systematically. A new kinematics method based on zero inclination orbit transformation was proposed. The influence of the reference satellite inclination on surrounding satellite orbit has been analyzed by the new kinematics method. According to the rules of remote sensing satellite design, a satellite formation with high coverage performance and almost same orbital altitude has been designed, and the accuracy of the direct georeferencing based on remote sensing image has been analyzed.The relative orientation and attitude determination of satellite-borne camera is a key problem of ground position. The performance of two linear algorithm for relative orientation were compared, and the results show that the linear algorithm based on SVD of essential matrix was not suit for relative orientation of stereo images, and the affine factorization method is worked well for relative orientation of multiple images. The relative orientation parameters by factorization method can been refined by the bundle adjustment. An attitude determination algorithm base on the base line of satellite formation was proposed, which provides the necessary camera attitude information to the direct georeferencing. Two dynamic attitude estimation algorithms based on satellite dynamic model were studied. The simulation experiment results show that the accuracy of attitude determination depends on the accuracy of relative orientation, and the attitude accuracy of camera can be improved by nonlinear predicitive filter.To improve height accuracy of direct georeferencing, the combined block adjustment with relative control was studied, and the results were compared with the block adjustment with a few control points. The methods of coordinates transformation from satellite network to earth network and space projection for map were introduced.Image feature detection and subpixel location were basic task of photogrammetry and remote sensing. Through the survey of Image feature methods, a junction detection method based multi-feature was proposed. The junction subpixel location error caused by image noise, feature scale and feature attribute was researched, and the subpixel location error of road featurecaused by gray difference of two sides was analyzed.The technique of area-based image matching was researched from reliability, accuracy and efficiency. Based on the needs of reliability and accuracy, a fast optimal matching area selection method of matching feature was presented. To reduce the search space, a feature point matching algorithm base moment invariant and singular value decomposition was proposed, and a search strategy based on improved genetic algorithms was proposed to improve the efficiency of image matching. The results show that the reliability, accuracy and efficiency of matching area based on region are improved by those methods and search strategies.更多还原