| 题名 | 高精度星敏感器处理链路关键技术研究 |
| 作者 | 姜亮 |
| 学位类别 | 博士 |
| 答辩日期 | 2015-05 |
| 授予单位 | 中国科学院大学 |
| 导师 | 任建岳 |
| 关键词 | 星点目标检测 SVD准最优匹配 导航星表创建 星图模拟 星图识别 |
| 其他题名 | Study on the Key Technologies of High-precision Star Sensor Processing Link |
| 学位专业 | 光学工程 |
| 中文摘要 | 星敏感器是一种以恒星为参照物的高精度姿态测量设备，它与地平仪、太阳敏感器、磁强计等传统姿态传感器相比，具有自主性好、抗干扰能力强、姿态测量精度高的优点。星敏感器已成为空间飞行器姿态测量系统中的核心设备，对空间遥感光学成像卫星来说，精确的相机视轴指向信息是完成高精度定向测绘和天基目标监视任务的有力保障。随着我国高分辨率光学遥感成像卫星技术的快速发展，国内星敏感器技术整体发展水平出现了无法与之匹配的问题。作为某型遥感相机研制课题的子课题，课题任务要求研制一款适用于高分辨率遥感光学成像卫星姿态测量的星敏感器，同时针对当前星敏感器高精度、低功耗、小型化的发展趋势，本文对星敏感器处理链路中的关键技术点展开研究，完成的主要工作包括： 以STAR1000(宇航级APSCMOS)为核心成像器件完成了星敏感器前端成像电路设计，图像以LVDS格式输出，实现了相机的4增益、16级曝光时间控制及起止式异步串口通讯，通过夜场成像测试验证了星敏感器相机功能。 研究了星点目标检测相关算法，文中提出采用边缘背景动态阈值方法对高斯滤波后的星图进行阈值分割，有效避免了硬阈值法带来的目标丢失和星点定位精度损失的问题。随后采用四连通域分析算法完成了星点目标标记，并选用带阈值的质心法进行了星点坐标提取，分析了定位误差的来源。文中提出采用改进的二维频域分析法对星点定位系统误差进行了理论推导，得到了以点扩散函数为参变量的系统误差理论解析式。并通过CODE V中导出的点扩散函数数据，对星点定位系统误差进行了仿真实验，实验结果与改进的频域分析结果相符，较好地解释了S曲线误差在图像坐标系平面内振幅改变的实验现象。通过星点定位系统误差测量与补偿实验对频域分析结果及误差补偿方程的补偿效果进行了验证，补偿效果优于传统正弦曲线补偿方法，星点亚像元定位精度达到1/40pixel量级。 以历元J2000下的史密松天文台星表(SAO)为数据来源，完成了导航星表的创建。首先基于极限星等对SAO星表中的约25万颗导航星进行筛选，采用天文校正算法完成了恒星的视位置转换。在天区划分方面，文中提出了一种以空间立体角为度量基准的均匀天区划分方法，解决了传统天区划分方法中天区划分不均匀的问题，并以该天区划分为基础对导航星分布进行了准均匀化处理，最终得到4096颗导航星构成的导航星集合。导航星表创建方面，文中提出了一种改进的导航星表存储结构，将导航星表分为相互配合使用的四个表(导航星信息表、星对表、角距快速查找表、导航星分区表),很好地配合了后续相关算法的执行，提高了各算法的信息检索效率。分析了恒星坐标由天球坐标系到像平面坐标系的转换，透视投影变换及非线性成像模型，最终通过星等到灰度的转换生成模拟星图，并在MFC框架下采用OpenCV图像视觉库完成了星图模拟软件设计，为后续算法地面仿真实验创造了必要的软件测试环境。 针对星敏感器全天识别模式，提出了一种基于星矢量矩阵SVD准最优匹配的星图识别算法，该算法采用星矢量矩阵奇异值分解(SVD)得到的特征向量三元素构造损失函数，并保留使损失函数最小的N个候选星组合作为准最优匹配，解决了子图同构类算法中冗余匹配过多且无法剔除的问题，提高了算法识别率。针对天球“空洞区”识别问题，提出了一种基于天球“空洞区”导航星查找表的识别方法，作为文中星图识别算法的一个备选补充分支。 姿态解算方面文中提出了一种改进的SVD姿态解算算法，该算法避免了传统SVD姿态解算算法中出现的输出多个姿态的问题，并且在姿态解算中利用星图识别过程的SVD中间数据用于姿态解算，减少了姿态解算算法时间开销。并通过星图模拟软件对TRIAD算法、QUEST算法、文中提出的改进SVD算法进行了对比实验，结果证明改进SVD算法具备与QUEST算法相当的精度，精度显著高于TRIAD算法。 针对星跟踪问题，采用了一种基于星图映射的帧间快速星跟踪算法。算法利用观测星图的帧间姿态信息，进行快速星图映射生成数字参考星图，以参考星图中的星坐标为中心进行邻域开窗，对星点目标实施跟踪。对于邻域半径内进入星不唯一的情况，引入角距匹配机制进行快速筛选，增强了传统星跟踪算法的跟踪能力。算法设置了星跟踪阈值，保证了星跟踪的可靠性，同时也保证了星跟踪模式与全天识别模式之间的合理切换。经仿真实验验证，算法具备在较大角速度机动条件下的稳定跟踪能力。 |
| 英文摘要 | Star sensor is a kind of high-precision attitude measuring instrument, which refers to stars on the celestial sphere. Compared with traditional attitude measuring instruments such as horizon sensor, sun sensor and magnetometer, star sensor has many advantages, which include better autonomy, stronger anti-interference ability and higher precision. Star sensor has become the key instrument of the spacecraft attitude measuring system. For the space optical remote sensing satellites, accurate direction of the boresight is an assurance of the high-precision direction-fixed mapping and space target surveillance mission. With the rapid development of high-resolution optical remote sensing satellite technology of our country, the development of the star sensor technology can not catch up with that pace. This project is the sub-project of a remote sensing camera project, of which the mission asks for a type of star sensor which can be used on the high-resolution optical remote sensing satellites. In the view of the trend of high-precision, low power consumption and miniaturization, the key technology points of star sensor processing link are studied in this dissertation. Taking STAR1000 as the core imaging device, the design of the imaging circuit is accomplished. The image data is output in LVDS format. Four-gain control and sixteen-grade exposure time control have been realized. The circuit can commu- nicate with other devices through a serial port. High quality of the star camera has been confirmed by the night sky test. The star targets detection algorithms are studied in the dissertation, and a dynamic threshold image dividing algorithm which gets the dynamic threshold from the edge background is put forward. The star image which has been dealt with by the Gauss filter is divided by this algorithm. The algorithm can avoid targets missing and star centroid precision decreasing. The star targets are labeled by four connected domain analysis algorithm, and then the star targets coordinates are obtained by the centroid algorithm with threshold. The sources of the centroid error have been analyzed. The star centroid systematic error is analyzed by the improved analysis method in two-dimension frequency domain. The theoretical expression of the star centroid systematic error has been obtained, of which the PSF is a parameter. A simulation experiment on the star centroid systematic error is carried out based on the PSF data from CODE V. The result of the simulation experiment is in accordance with the frequency analysis, which can explain the experiment phenomenon that the amplitude of the S curve changes in the image plane. Through the star centroid syste- matic error measurement and compensation experiment, the result of the frequency analysis and the effect of the error compensation equation have been validated. The effect of the error compensation equation is better than the traditional sinusoidal curve compensation method, and after the compensation the star sub-pixel centroid accuracy reaches about 1/40 pixel. With data from the Smithsonian observatory catalog (SAO) under epoch J2000, the establishment of the guide star catalog has been accomplished. Firstly, based on the limit magnitude guide stars are selected from about 250,000 stars of the SAO, and then the conversion of star apparent position is carried out by astronomical correction algorithm. For the problem of celestial sphere division, this dissertation proposes a uniform celestial sphere division method based on the space solid angle. The method solves the problem of uneven division which the traditional methods have encountered. Based on the celestial sphere division of this dissertation, a quasi-uniform treatment to the guide stars is carried out, and 4096 guide stars are obtained finally. The guide star catalog is established with an improved storage structure, including four tables which are star angular distance information table, star pair table, star angular distance lookup table and the celestial sphere division table. These four tables cooperate with each other and improve the efficiency of infor- mation retrieval. The four tables cooperate well with the following algorithms. The analysis of star coordinates transformation from the celestial coordinate system to the image plane coordinate system is carried out. And the projection transformation and nonlinear imaging model are also discussed in this dissertation. After the gray conversion the simulation star map is generated. Finally under the framework of MFC the design of star map simulation software is completed using OpenCV vision database, establishing a software testing environment which is necessary for the simulation experiments for following algorithms. In this dissertation, a star map identifying algorithm based on star vector matrix SVD quasi-optimal matching is proposed. The algorithm uses the three elements of the feature vector obtained by the singular value decomposition of the star vector matrix to construct the loss function, and keep the last N candidate star groups which generate the N smallest loss function values. The algorithm solves the redundancy problem encountered by the sub-graph isomorphism algorithms, improving the identification rate of the algorithm. For the identification of the celestial regions which get few stars, an identification method is proposed based on the "empty region" guide star lookup table, which is an alternative branch of the identification algorithm. On the aspect of attitude algorithm, a kind of improved SVD attitude algorithm is proposed. The algorithm solves the problem that the traditional SVD algorithm may output two attitudes for one frame. The improved SVD attitude algorithm uses the data generated by the identification algorithm, which reduces time consumption. Through the simulation software of star map, the contrasting experiment of the improved SVD algorithm, QUEST algorithm and TRIAD algorithm is carried out in the dissertation. The result shows that the improved SVD algorithm is as precise as QUEST algorithm, whose accuracy is significantly higher than that of TRIAD algorithm. On the aspect of star tracking, an algorithm based on fast star image mapping is used. The algorithm uses the attitude information among frames to generate digital reference star image by star image mapping method. Taking the star coordinates in the reference star image as the centers of tracking windows, the star tracking is carried out. If one star is not the only one in a tracking window, the angular distance matching mechanism will help to identify that star, which enhances the tracking ability of the algorithm. A threshold is set to ensure the reliability of the star tracking algorithm. The threshold also ensures the reasonable switching between the star tracking mode and the celestial identification mode. The result of the simulation experiment indicates that the tracking algorithm has the ability to track stars steadily under a large angular speed condition. |
| 公开日期 | 2015-12-24 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ciomp.ac.cn/handle/181722/48848]  |
| 专题 | 长春光学精密机械与物理研究所_中科院长春光机所知识产出 |
| 推荐引用方式 GB/T 7714 | 姜亮. 高精度星敏感器处理链路关键技术研究[D]. 中国科学院大学. 2015. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论