摘要
针对探月飞行器在接近、下降、着陆阶段所要完成的精确导航任务,提出了一种基于多尺度光流法进行探月飞行器的自主视觉导航的方法.在探月飞行器的下落过程中,由于与探月飞行器固联CCD所拍摄的图像在相邻时刻之间的相对运动较大,因而传统的光流计算方法无法同时满足精度和鲁棒性的要求.本文引入了一种在图像小波金字塔的基础上多尺度光流法,在图像间相对运动较大的情况下实现了对特征点的跟踪.最后,利用鲁棒最小二乘方法,估计出探测器在下落过程中不同时刻之间的相对刚体运动参数.仿真结果表明该方法快速、有效的完成探月飞行器的自主视觉导航任务.
Aimed at the lunar lander's precise navigation task during the entry,descent and landing(EDL) phase,a vision navigation method using multi-scale optical flow is proposed.During the descent phase,there exists large relative motion in the image sequence captured by the CCD camera mounted on the lander.The original optical flow method can not satisfy the requirement in both accuracy and robustness.An image wavelet pyramid based multi-scale optical flow is introduced to make the features can be tracked between images with large motion.By using the robust least square method,the parameters of the relative rigid motion can be precisely estimated.The simulation results show that the vision navigation method is fast and effective for the autonomous navigation task during the EDL phase.
出处
《传感技术学报》
CAS
CSCD
北大核心
2007年第11期2508-2512,共5页
Chinese Journal of Sensors and Actuators
基金
国家自然科学基金重点资助项目(60475027)
关键词
视觉导航
多尺度光流法
运动估计
vision navigation
multi-scale optical flow
motion estimation
