【作者】 傅丹；

【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， 2008， 博士

【摘要】 本文以空间平台对空间目标采用光测图像方法进行观测为背景,研究了基于直线特征重建空间非合作目标三维结构和测量空间合作目标分离后与空间平台相对位姿的有关算法。空间环境中存在大量不可控制或者不可预知因素,执行空间任务时摄像机系统的参数容易受到影响,因而在轨对摄像机参数进行标定具有实用意义。同时,大部分空间目标包含丰富的直线特征信息。本文充分利用直线特征的性质,展开了以下研究工作。1、空间非合作目标提供的目标特征信息有限,不能简单采用合作目标的光测图像方法进行观测。由于直线之间存在平行、垂直和相交等相互关系,以基于直线特征的中心透视投影方程来描述摄像机成像几何模型,以场景中直线的几何关系作为算法约束,针对单幅图、图像对和序列图像的情况,开发了标定摄像机内外参数并重建空间非合作目标三维结构的新算法。这是全文的重点。1)关于未标定的单幅图,提出了一种标定摄像机内参数的新方法,该方法利用了直线的射影不变性中的平行直线相交以及直线的相似不变性中的正交直线在相似变换中保持正交两个性质。在此基础上,基于直线的仿射不变性中的平行直线在仿射空间保持平行以及直线的相似不变性中的虚圆点在相似变换下保持不变的性质重建平面结构。定量分析了摄像机参数选择对消影点精度的影响,研究了如何合理选择摄像机的拍摄位置和角度来高精度求取消影点。针对近景测量像差畸变大的情况,讨论了基于直线的射影不变性来纠正鱼眼镜头像差畸变的方法。2)关于未标定的图像对,提出了基于直线的射影不变性中的交比不变和极线约束标定摄像机内外参数并重建目标三维结构的算法。3)关于未标定的序列图像,提出了基于直线的相似不变性中的正交直线在相似变换中保持正交和光束法平差标定摄像机内外参数并重建目标三维结构的算法。针对本文研究的空间非合作目标,设计仿真实验研究了空间平台与非合作卫星非共面一般接近以及空间平台绕飞非合作目标时,分别采用基于图像对和序列图像的方法来重建非合作目标三维结构的问题。2、本论文提出了采用固连主副摄像机的测量方案,基于直线特征测量合作目标分离后与空间平台的相对位姿的方法。该方法分别使用主摄像机和与之固连的副摄像机实时拍摄分离系统和空间平台上的基准标定物的图像;通过提取图像中的直线特征,利用合作目标中已知的空间直线精确位置关系,推导分离目标相对合作平台位姿的解算公式,获得合作目标与空间平台的相对位姿。3、作为图像分析的预处理步骤,研究了直线特征的提取和匹配算法。1)提出了一种最大梯度方向Hough变换与一维高斯模板拟合相结合提取直线的两步算法,既合理利用了已有的经典算法,又提高了直线提取的精度。2)针对三维重建的需求,提出了一种基于极线约束和RANSAC算法的直线段匹配方法,有效解决了图像中被部分遮挡直线段的匹配问题。数字仿真和地面实验表明,本文提出的这些采用光测图像的算法对约束条件要求低,易于实现,抗噪性强,具有较高的精度,为解决空间目标观测中的摄像机参数标定、目标位姿解算和结构重建提供了若干种基于目标直线特征的实用算法。更多还原

【Abstract】 In this dissertation, the methods and algorithms based on line feature are studied, by using which 3D structure of non-cooperative targets is recovered and relative pose between separated cooperative targets and the space platform is measured. The background of the works is to observe space targets on the space platform with the images of targets and scenes captured by cameras.It’s useful to calibrate camera parameters on-line for the camera parameters tend to be disturbed by the unpredictable factors in the harsh environment. In the meantime, there are plenty of line features in the most of the space targets, so the main contents of the works in the dissertation are done as follows by using the characteristics of line sufficiently.1 There is little 3D information in non-cooperative targets so that 3D structure could not be reconstructed with the methods of cooperative targets simply.The algorithms used to measure the cooperative targets can not be simply used to measure the non-cooperative targets for they can only provide limited features.There are correlations of parallel, vertical and intersection between the straight lines, so the geometric model of camera is described with the central perspective equations based on the straight line feature and geometric constraint of the straight lines in the scene is used. The methods of camera self-calibration and 3D structure reconstruction from non-cooperative targets images, including single image, image pairs and sequential images, based on line feature are introduced in detail. This is the emphasis of the article.(1) An algorithm is proposed from uncalibrated single image to calibrate the camera inside parameters, based on the projective invariance that parallel lines should be intersect at one point and the similarity invariance that the vertical lines remain vertical. Then, the method to reconstruct the plane is introduced when the inside parameters of camera is calibrated, based on the affine invariance that parallel lines remain parallel and the similarity invariance that the circular points are fixed points.The effect about the choice of the camera parameters on the precision of vanishing point is analyzed quantitatively and how to choose the location and the angle of camera parameters reasonably to get the vanishing point of high precision is discussed. A method to correct the fish-eye aberration distortion, which was based on the projective invariance that straight line remains straight, was introduced to the situation of aberration distortion being deteriorated in close-range measuring.(2) An algorithm is put forward from uncalibrated image pairs to calibrate the parameters of camera and reconstruct the 3D structure, based on the projective invariance that cross-ratio should be constant and epipolar line constraint.(3) An algorithm is presented from uncalibrated image sequence to calibrate the parameters of camera and reconstruct the 3D structure, based on the similarity invariance that vertical lines remain vertical and bundle adjustment.In order to study non-cooperative targets, the simulation experiments are conducted. The algorithm of image pairs is applied when the space platform and the non-cooperative target are generally approaching. The algorithm of image sequence is used when the space platform flies around the non-cooperative target.2 The relative pose measurement methods using the system of fixed main camera and accessorial camera are applied.The images of the space separating targets and the reference calibrated target on the space platform are captured in real-time by the main camera and accessorial camera respectively. Relative pose between the cooperative targets and the space platform is computed provided that the lines of the images captured by the cameras are extracted and the location of 3D lines in cooperative target is known accurately.3 As steps of image pre-processing, the straight line feature extraction and the match algorithm are studied.(1) A two-step algorithm of the straight line extraction based on the Hough transform with the gradient’s maximum direction and 1D Gaussian templet fitting is proposed. It not only reasonably uses the existed classical algorithms, but also improves the accuracy of the straight line extraction.(2) A straight-line-segment match method based on the epipolar line constraint and RANSAC algorithm is applied with regard to the demand of the 3D reconstruction, which effectively solved the matching problem of sheltering straight line segment in the image.The algorithms presented in this dissertation are all based on line feature. They have effectively solved the problems of camera self-calibration, space targets’ 3D structure and pose measurements. Their precision and efficiency are verified by a lot of digital and ground synthetic experiments.更多还原