【作者】 张昊；

【导师】 郁道银； 黄战华；

【作者基本信息】 天津大学 ， 光学工程， 2005， 博士

【摘要】 某兵器实验基地使用高速电视光测系统测量多种导弹和无人机的姿态。姿态参数包括空间姿态和单站姿态。空间姿态是指中轴线的俯仰角和偏航角,测量误差不超过1o;单站姿态是光测系统拍摄的单站高速视频中目标平面中轴线方程,设计要求平面中轴线的倾角误差不大于0.5o,截距误差不大于0.5个像素。本文的主要任务是设计事后姿态判读软件,实现单站姿态的自动判读,再利用“中轴法”交汇计算空间姿态。本文的主要创新点包括:1、提出通过计算无人机和导弹的二维运动参数,实现单站姿态自动判读。先手工建立初始时刻的目标中轴线方程,再利用运动视觉分析的方法计算目标的运动参数,进而推算出后继时刻的目标中轴方程。2、通过光流分析方法计算目标运动参数。提出基于角点跟踪的运动参数鲁棒估计算法。利用基于卡尔曼滤波的KLT角点跟踪算法,得到相邻图像间物体上角点的对应关系。再通过RANSAC算法排除错误的角点匹配,并利用余下的角点对应关系,使用最小二乘法计算运动参数。3、用贝叶斯滤波方法求解运动参数估计问题。提出基于Condensation算法的运动参数估计方法。以运动参数为测量向量,当前时刻和上一时刻的运动参数为系统状态向量。对初始段的无人机,建立线性的系统方程和测量方程。使用轮廓模型与边缘点的相似性测度来评价状态向量的后验概率,并通过距离变换提高计算效率。4、利用轮廓配准算法求运动参数。提出“Condensation+DE(差分进化)”运动参数估计算法。将Condensation算法的“粒子”分布空间作为轮廓配准的搜索空间,使用差分进化法求解轮廓配准中的全局最优化问题。从另一个角度,也可以认为该方法将Condensation的状态向量后验估计问题转化成一个最优化问题,并使用差分进化法求解,可以克服Condensation的退化问题,降低对系统方程准确性的要求,提高计算的可靠性和精度。本文的研究成果-目标姿态自动判读软件,已通过有关单位验收并交付使用。在实际应用中取得了满意的效果。更多还原

【Abstract】 A digital high speed video camera system is utilized by an ordnance test center tomeasure attitude of various kinds of missiles or UAVs (Unmanned Aerial Vehicle).Attitude parameters comprise 3D attitude and substation attitude. 3D attitude refers tothe pitch and yaw angles of the axis, with a measure error of no more than 1o. Andsubstation attitude is the line equation of axis in the monocular highspeed video shotfrom substation. The error of the slope is no more than 0.5o and that of the interceptdoes not exceed 0.5 pixel. The intention of our research is to develop a post-flight data processing softwarethat can auto-interpret the substation attitude and compute the 3D attitude accordingto its substation attitude by “axis joint method”. The major innovations of this dissertation are as follows: 1. The auto-interpretation of substation attitude is mapped to a framework of theestimation of 2D motion parameters. The intuition behind this idea is that if themotion parameters of a target are computed by the method of motion analysis and theline equation of axis in the first frame is known, the substation attitude in thosesequential frames can be estimated. 2. A robust motion estimation method based on corner tracking is proposed. AKalman-based KLT corner tracker is proposed to match the corners on a targetbetween neighboring frames. A robust estimator, RANSAC is utilized to exclude falsecorrespondences. And according to those correct correspondences, systems of linearequation are constructed and solved by the method of least squares. 3. A Condensation-based motion estimation method is proposed. In theframework of Condensation, motion parameters are taken as the measure vector,while the state vector is composed of current time motion parameters and those of lasttime. Linear system equation and linear measure equation that describe the attackingUAV are modeled. Post density of the state vector is evaluated by shape similaritybetween the contour model and edge points, and distance transformation is introducedto improve calculation efficiency. 4. A motion estimation method based on “Condensation+DE (DifferentialEvolution)” is proposed. The parameter distribution space of Condensation is taken asthe search space of contour registration, while DE is utilized as the global optimizer.From the other view, the estimation of posterior density in Condensation can beregarded as the problem of global optimization solved by DE. With the importation ofglobal optimization, degeneracy problem of Condensation is avoided, demand isreduced on accuracy of system model, and estimation of motion parameters is morerobust and accurate.As a result of our research, the attitude auto-interpretation software has beentested and accepted by the ordnance test center.更多还原