基于视觉序列图像的月球车自运动估计技术

Ego-motion Estimation of Lunar Rover Based on Vision Sequence Images

月球车视觉所获得的环境信息是月球车进行导航的重要信息来源,给出了一套利用月球车立体视觉系统所获得的序列图像来估计月球车运动参数的定位方案,并重点解决了后期的定位问题。该方案首先通过特征提取、特征跟踪与特征匹配等图像处理环节建立运动估计所需要的特征匹配点集,然后针对运动参数估计这一难点,利用鲁棒运动估计方法有效消除匹配特征点集中的误匹配,并初步估计月球车运动参数,以此为迭代初值利用Levenberg-Marquardt非线性估计算法精确估计月球车运动参数,最后通过建立运动估计仿真器验证了该运动估计方案的有效性。

Lunar environment information that coming from lunar rover stereo vision system is a very important source for lunar rover. This paper presents an motion estimation algorithm that will enable lunar rover travels in unstructured lunar environment autonomously, our techniques are based on the stereo vision sequence images, firstly, matched-features aggregate are build by image processing step such as feature detection, feature tracking and feature matching, next, robust motion estimation is executed to estimate the motion initially and reject the outliers, and then, Levenberg-Marquardt nonlinear estimation is executed to estimate the motion precisely, the output of our algorithm is motion parameters （attitude and localization） of lunar rover, finally we testified its validity by simulation.