基于天体三维模型的着陆段视觉导航方法

A visual navigation algorithm for landing based on 3D model of small body

针对深空探测精确着陆的需求,本文提出一种基于天体三维模型的着陆器视觉导航方法.该方法以天体三维模型为数据基础,根据着陆器初始猜测状态在线生成导航地图,通过已知路标匹配确定着陆器相对着陆点位置姿态;借助三维模型提供的深度信息配准像素灰度,恢复序列图像帧间运动,给出两次已知路标匹配之间的局部运动轨迹,实现着陆下降过程着陆器状态的准确估计.在线生成参考地图与实际采集的天体图像在尺度、视角等方面比较接近,更易于导航路标的匹配,解决了下降过程尺度大范围变化使路标难以匹配的问题.引入像素深度信息辅助灰度配准,通过搜索图像间平移和旋转使图像中的全部特征灰度差异最小,采用了全局优化思想,因此可以克服特征匹配过程易陷入局部极小,出现误匹配的问题,具有更高的帧间运动估计精度和稳定性.数学仿真表明,本文提出的视觉导航方法具有良好的性能,满足深空探测高精度着陆的导航需求.

A visual navigation algorithm for landing based on 3 D model of small body is propoesd to address the requirements of precise landing in deep space. This approach generates reference image and calculates depth of pixels online according to the guess of lander state as well as 3 D model of small body, and estimates lander pose relative to the final landing area through matching landmarks of reference image and realistic image collected from navigation camera. Then, we estimate motion of lander with sequential images by aligning pixel intensity of images between two matching events, to achieve accurate navigation for lander. The approach addresses the challenge of large different scale when lander descends from hundreds of thousands meters to meters, which results in higher landmark match rate and more precise motion estimation. In addition, pixel intensity alignment adopts global constraint of gray image, optimizing the translation and rotation between two images until the residual of intensity is minimized, which is able to overcome local minimum problem of motion estimation and improve the robustness, stability of algorithm. Finally, the simulation results show that this visual navigation method performs well and could attain requirements of high precision landing in deep space probe.