基于特征跟踪辅助的小行星着陆先验位姿纠正

Prior Pose Correction for Asteroid Landing Based on Feature Tracking Assistance

针对小行星着陆视觉导航过程中可能出现的先验位姿不准确场景,提出一种基于特征跟踪辅助的位姿估计方法。利用先验位姿信息和三维导航特征数据库实时生成导航特征,并利用手工与深度特征联合的多特征判别式相关滤波跟踪导航特征在导航相机影像中的位置,进一步利用平均峰相关能量筛选出可靠的跟踪结果用于位姿初步估计,基于该初步估计的位姿重新生成导航特征,并使其与导航相机影像做归一化相关匹配,该过程将在可微分的列文伯格−马夸尔特框架中基于归一化相关系数约束来实现位姿优化。利用“冥王号”(Origins,Spectral Interpretation,Resource Identification,and Security-Regolith Explorer,OSIRIS-REx)任务的影像、地形和星历数据进行位姿估算实验,结果表明:在先验位姿不同程度噪声的干扰下,提出算法的位姿估计重投影误差均为亚像素,在距离小行星表面1 km处,位置估算误差在2 m以内,姿态估算误差在1°以内,可有效纠正不准确的先验位姿。

Aiming at the possible inaccuracy of the prior pose during the visual navigation of an asteroid landing,a feature tracking aided pose estimation method is proposed.First,the generation of navigation features relies on pre-existing pose information and a database of navigation features.Subsequently,a multi-feature discriminative correlation filter(DCF)is employed to track the position of the navigation features in the navigation camera images by combining handcrafted and depth features.The average peak correlation energy(APCE)is subsequently employed to effectively screen dependable tracking outcomes for the initial estimation of the pose.Finally,the navigation features are recalculated using the initial estimation of the pose and adjusted to match with the navigation camera image by using normalized correlation coefficients(NCC).The proposed methodology involves the integration of the process within a differentiable Levenberg-Marquardt(LM)framework,specifically designed for pose optimization.This framework incorporates constraints based on the NCC.Experimental results,utilizing images,terrain,and ephemeris data obtained from the Osiris mission,demonstrate that the proposed method's pose estimation exhibits reprojection errors within the sub-pixel range.At 1 km from the asteroid surface,the position estimation error is within 2 m and the attitude estimation error is within 1°.