基于摄站点云的巡视器相对定位方法

A Relative Positioning Method for Rover Based on Fusion of Camera Station Clouds

针对传统特征点法巡视器相对定位受摄站间距离及影像视角变化影响,导致相对定位失效的情况,结合着陆器降落序列影像及巡视器导航影像提出了一种基于点云配准的巡视器相对定位方法。首先,依托降落序列影像恢复着陆区三维点云作为基准底图,并获取着陆点初始位置;其次,利用附加残差模块的多尺度代价聚合立体匹配方法对巡视器摄站区域进行精密三维重建;最后,结合着陆区三维点云与摄站点云利用SAC-IA+ICP方法进行点云配准,并通过RANSAC获取可靠的巡视器相对定位参数。开展了嫦娥四号巡视器相对定位实验。实验表明,所提出的方法较传统光束法平差定位结果在x方向和y方向平均误差优于0.56 m和0.52 m,最大误差优于0.11 m,表明该方法可为巡视器智能感知与长距离导航定位提供一定的参考。

In response to the situation where the relative positioning of traditional feature point method inspectors is affected by the distance between rover camera stations and changes in image perspective,resulting in the failure of relative positioning of the rover,this article proposes a relative positioning method based on point cloud registration for rover,combining landing sequence images of landers and navigation images of rovers.Firstly,based on the landing sequence images,the 3D point cloud of the landing area is restored as the baseline map,and the initial position of the landing point is obtained.Secondly,the multi-scale cost aggregation stereo matching method with additional residual modules is used to perform precise 3D reconstruction of the camera station area of the rover.Finally,combining the 3D point cloud of the landing area with the camera station cloud,the SAC-IA+ICP method is used for point cloud registration,and reliable relative positioning parameters of the inspector are obtained through RANSAC.It conducts relative positioning experiments on the Chang’e-4 rover,and the results show that the proposed method has an average error of 0.56 m and 0.52 m in the x and y directions,and a maximum error of 0.11 m compared to the traditional bundle-adjustment positioning method.This indicates that the proposed method can provide valuable reference for intelligent perception and long-distance navigation positioning of the rover.