基于计算机视觉的探测器运动估计算法研究.pdfVIP

2008 年航空宇航科学与技术全国博士生学术论坛 2008 年 10 月 基于计算机视觉的探测器运动估计算法研究 田阳 崔平远 崔祜涛 (哈尔滨工业大学深空探测基础研究中心，哈尔滨 150080) 摘 要: 本文给出了一种深空探测器着陆阶段的自主运动估计算法。该算法首先对导航相机采集的图像序列进行 预处理，通过 Shi-Tomasi-Kanade 算法检测与跟踪相邻两幅图像中多个对应特征点，得到匹配的特征点对。根据 对极约束原理，采用线性算法给出基础矩阵的初始值，由于线形算法的简化使这一初始值精度难以满足导航要求， 且考虑到特征点匹配过程中会出现误匹配问题，采用鲁棒方法和非线性算法对基础矩阵进行计算，得到更为精确 的基础矩阵估计，结合激光测距仪给出的距离信息得到三维运动参数。因此，结合探测器开始进行导航拍照时在 着陆点坐标系下的位姿初始信息，可以给出探测器在着陆阶段相对该坐标系的全部运动信息。仿真结果表明，该 算法满足深空探测任务着陆阶段对导航算法自主性、实时性和可靠性要求。 关键词: 小天体，软着陆，运动估计，计算机视觉，特征点跟踪 中图分类号: V448.22 文献标识码: A A Study on Deep Space Explorer Motion Estimation Based on Computer Vision TIAN Yang, CUI Ping-Yuan, CUI Hu-Tao (Deep Space Exploration Research Center, Harbin Institute of Technology, Harbin 150080, China) Abstract: This paper proposes an autonomous motion estimation method for soft landing on small body. The estimation algorithm of relative pose and motion from feature points tracked through an image sequence presented here is based on Epipolar Geometry. First the image sequence is processed to detect and track the feature points by Shi-Tomasi-Kanade algorithm. The initial value of fundamental matrix is computed by the robust linear motion algorithm as the Least Median of Squares method. Then the nonlinear motion estimation algorithm which is attempts to minimize the image plane error between the features in the two views is employed to obtain the 5DOF motion parameters. The Laser altimeter data is used to compute the scale of the motion parameters. The results on synthetic image show that the proposed algorithm can provide the estimation of state with satisfactory accuracy. Key words: Small body; Soft landing; Motion Estimation; Comp