【作者】 侯敏；

【导师】 张泽旭；

【作者基本信息】 哈尔滨工业大学 ， 机械（专业学位）， 2022， 硕士

【摘要】 随着空间在轨服务能力的提升,对空间非合作目标在轨操作和处理的任务越来越多。利用航天器在轨载荷与处理设备快速精确地测量出空间非合作目标的运动特性参数是交会导航的基础,对在轨服务与操作具有重要的价值。本文以此为研究背景,利用相机载荷采集的图像序列,实现对空间非合作目标相对姿态和运动参数的估计。本文主要的研究工作如下:首先,对空间非合作目标进行识别,研究一种基于FPN与Darknet网络进行空间非合作目标的识别方法,有效改善图像背景噪声对目标特征点检测带来的干扰。针对FPN网络存在对特征图中的信息利用率偏低的问题,研究基于不同尺度的空间非合作目标识别优化方法,使用残差增强的方法来补充顶层特征图中缺失的信息,并采用自适应特征融合网络,使得特征图上的信息更加丰富,达到了在不同尺度下对空间非合作目标的准确识别。其次,对空间非合作目标进行位姿估计,研究一种基于序列图像的空间非合作目标相对位姿估计算法。对初始两帧图像使用五点法解算相机的外参,基于三角测量的方法进行初始化,后续新增添的图像使用EPNP算法估计空间非合作目标的相对位姿,最后使用BA优化和三角化的方法提高相对位姿的解算精度,并通过全局BA优化算法消除误差的漂移。第三,对空间非合作目标进行运动特性估计,研究一种基于序列图像的空间非合作目标运动特性参数的估计算法。计算目标本体坐标系相对相机坐标系的相对姿态,利用空间非合作目标的姿态运动学方程和动力学方程,设计了一组扩展卡尔曼滤波器,估计空间非合作目标的角速度、惯量比及章动角等运动特性参数。最后,对本文提出的基于序列图像的空间非合作目标的运动特性估计方法进行了仿真实验,对本文提出的算法进行实验分析与验证。更多还原

【Abstract】 With the improvement of space on orbit service capability,there are more and more on orbit operations and processing tasks for non-cooperative space targets.It is the basis of rendezvous navigation to measure the motion characteristic parameters of non-cooperative space targets quickly and accurately by using spacecraft on orbit loads and processing equipment,which is of great value for on orbit service and operation.Taking this as the research background,this paper uses the image sequence collected by the camera load to estimate the relative attitude and motion parameters of the non-cooperative space targets.The main research work of this paper is as follows:Firstly,the non-cooperative space target is recognized,and a recognition method of non-cooperative space target based on FPN and Darknet network is studied to effectively improve the interference of image background noise on target feature point detection.In view of the low utilization of information in the feature map in FPN network,the optimization method of non-cooperative space target recognition based on different scales is studied,the residual enhancement method is used to supplement the missing information in the top-level feature map,and the adaptive feature fusion network is used to enrich the information on the feature map and achieve the accurate recognition of non-cooperative space targets at different scales.Secondly,a relative pose estimation algorithm of non-cooperative space targets based on image sequences is studied.For the initial two frames of images,the five point method is used to solve the external parameters of the camera,and the triangulation method is used to initialize.For the subsequent newly added images,the EPNP algorithm is used to estimate the relative pose of the non-cooperative space target.Finally,the BA optimization and triangulation methods are used to improve the accuracy of the relative pose,and the error drift is eliminated through the global BA optimization algorithm.Thirdly,we estimate the motion characteristics of non-cooperative space targets,and study an algorithm for estimating the motion characteristics of non-cooperative space targets based on image sequences.The relative attitude of the target body coordinate system relative to the camera coordinate system is calculated.Using the attitude kinematics and dynamics equations of the space non cooperative target,a set of extended Kalman filter is designed to estimate the angular velocity,inertia ratio and nutation angle of the non-cooperative space target.Finally,simulation experiments are carried out on the proposed method for estimating the motion characteristics of non-cooperative space targets based on image sequences,and the proposed algorithm is analyzed and verified by experiments.更多还原