【作者】 李蒙；

【导师】 谌德荣；

【作者基本信息】 北京理工大学 ， 兵器科学与技术， 2016， 博士

【摘要】 论文系统地研究了锥形诱饵飞行位姿视觉测量中的图像处理问题。通过对标记设计方法、快速直线检测算法、高精度位姿解算算法的研究,解决了锥形诱饵图像特征信息的有效性、图像特征信息的快速检测、位姿参数的高精度解算等问题,为锥形诱饵飞行位姿高精度视觉测量系统的工程实现提供了技术支撑。论文主要研究内容及成果包括:(1)针对锥形诱饵有效特征信息不足的问题,提出最优动态窗口标记设计方法并设计诱饵标记。在诱饵成像区域最小的极限姿态下确定最优动态窗口,在最优动态窗口中设计完备的位姿解算特征,解决飞行姿态变化导致的特征信息遮挡问题;以特征尺度、特征间距最大化为标记设计原则,选择锥形诱饵的母线作为特征要素,解决测量距离变大(特征尺度变小)导致的特征信号信噪比低的问题。仿真测试结果验证了锥形诱饵标记的有效性。(2)针对直线特征的弹上实时检测问题,提出优选候选点的快速直线检测算法。算法利用间隔扫描及非极大值抑制优选候选点,大幅减少参与直线生长等操作的边缘点数量,有效降低直线生长的计算量;针对候选点离散化(优选导致)导致的无法直接生长成直线的问题,算法利用梯度方向一致性准则、NFA准则及通路准则组成联合匹配准则判断相邻候选点的共线性,快速搜索出同一直线上的所有候选点并生成完整直线。仿真测试表明,提出算法计算量小且直线检测率高。(3)针对诱饵位姿高精度解算的问题,提出基于直线点加权的高精度位姿解算算法。算法利用直线点直接建立目标函数解算位姿,而非常规地先拟合直线后基于直线解算位姿,消除直线拟合过程的误差;根据直线点到所属直线解析平面的距离(与直线点精度成反比)设计直线点的权重系数,通过给距离远的直线点赋予小的权重系数降低直线点误差影响,提高姿态解算精度及抗噪性。针对利用交于一点直线求解位姿时存在的位移向量有无穷解的问题,算法利用点线结合的方式计算位移向量的唯一解。仿真测试表明,提出算法位姿解算精度高于典型算法。(4)针对常规目标飞行位姿视觉测量中的图像处理问题,论文还研究了:(a)提出基于变量约束的高精度位姿解算算法,算法利用三维直线点、二维直线及摄像机光心间的共面关系(约束能力强)构建方程组,使直线特征信息得到充分利用;根据待求变量(向量)各分量之间的不变性约束关系,在待求变量解空间的多个维数上计算位姿参数,选择最优值作为解算结果,解决现有算法假定解空间维数为1导致算法精度低的问题;(b)提出基于变量约束的点线结合位姿解算算法,算法联合点特征与直线特征构建方程组解算目标位姿参数,提高算法对目标特征信息的利用率。仿真测试表明,两算法精度及稳定性均高于典型算法。更多还原

【Abstract】 Bait fighting pose estimation was very important to bait micro-motion analysis,bait optimal design,or even the missile penetration capability.The paper is focusing on the reality demands of measuring the relative pose parameters between the bait and carrier with high-speed cameras,and researching on the image processing technology for bait high-precision pose estimation.The research contents include designing marker on the curved surface of the bait,researching on fast line detection method,and studying on high-precision pose estimation method,which are aim to provide technical support for the engineering implementation of the bait fighting pose estimation system.The paper also studied the pose estimation problem of conventional targets,which makes the results of the paper more widely used.The research results of the paper are as follows:(1)Designed a bait marker to provide feature information and summarized marker designing methods for rotary targets.The optimal dynamic window was determined under the extreme attitude to ensure that the camera can always capture a full window,and sufficient number of features was put in the window to solve the problem of partial occlusion.According to the structure characteristics of the dynamic window,straight lines which are both stable and scale invariant were chosen as marker features and the length of lines were maximized to make it easy to detect.Marker designing methods for rotary targets were summarized based the bait marker designing experiences.Simulation results show the effectiveness of the bait marker.(2)Proposed a candidate points chosen line detection algorithm for real-time feature detection.The proposed algorithm uses selective scanning technique together with non-maximal suppression techniques to significantly reduce the number of edge points needed to perform line growth.The reserved candidate points are discretized and cannot directly grow into a straight line.Due to which,the gradient direction consistency criterion,NFA criterion and path criterion were used to consist of the joint-matching criterion.Using this criterion,the adjacent candidate points are matching to generate line kernels and the line kernels rapidly grow into complete lines.Simulation tests show that the proposed algorithm is of low computation and high line detection rate.(3)Proposed a weighted pose estimation algorithm to obtain bait’s high-precision pose parameters.The unrelated points on 2D lines are used to construct the objective function,basing on which the fitting process of 2D lines are avoided.A weight is allocated to each 2D line point based on the distance of the point to its corresponding projecting plane,with which the contribution of the point to the objective function are precise controlled,because of which the line features are used more effectively.At last,the translation is solved linearly by introducing several point correspondences.Experiments show that,the propose method outperforms other methods in term of accuracy and noise robustness.(4)Proposed an accurate and stable pose estimation algorithm by introducing the unchanged information contained in unknowns.To use the line constraints more effectively,equation system is constructed based on the points of 3D lines,as the constraints of these points are stronger than that of line direction vectors.To obtain noise insensitive solution,the equation system is solved in several possible dimensions by introducing the unchanged constraints contained in unknowns,and the pose parameters with the smallest error are chosen as the final result.On the basis of this algorithm,a more accurate pose estimation algorithm that can unit line and point features were proposed.The 3D point and line features are represented by one control point and three control vectors,and the constraint equations constructed basing on these features are combined with a reasonable weight.Then the equations are solved to get the pose parameters by introducing the unchanged constraints contained in unknowns again.Experiments show that,the two algorithms were both more accurate and stable than other non-iterative methods.更多还原