玉兔二号月面大间距变倾角成像的空间分辨率建模及其应用分析

Modeling and application analysis of the pixel-spatial resolution of Yutu 2 rover with large-scale transformed images

玉兔二号月球车在月面巡视探测通过地面遥操作规划控制完成.为了最大化每次规划的行驶距离,每次行进通常在6~9 m之间,使得相邻导航站点距离较大,所拍摄图像的尺度、旋转、平移上有较大的变化,图像重叠度较低且重叠区分辨率与形态差异很大,为站点间图像的自动匹配带来极大困难,进而影响相邻站点视觉定位的自动化进程.针对此问题,本文在前期关于图像像素空间分辨率建模的基础上,结合任务实际新增月面凹凸倾斜等因素对像素空间分辨率的影响,建立了更为通用的像素空间分辨率模型.在此基础上,推导了不同站点距离、成像距离和月面倾角等因素影响像素空间分辨率的等效关系,确立了两两因素影响分析的最少组合.最后结合玉兔二号第一、二月昼间所有的站点数据,分析了像素空间分辨率比值对图像匹配成功率的影响关系,对不同类型的影响情况进行分类,总结了不同因素对图像匹配成功率影响的一般性规律,给出了实际任务中视觉定位对相邻导航点相对距离和位置选择的约束准则,这对巡视探测任务中导航站点选择规则的制定和视觉自动定位能力的提升具有重要的指导意义.

In Chang’E-4 lunar exploration mission,Yutu 2 rover tasks are planned and controlled by teleoperation.The rover usually travels between 6 and 9 m to cover a large distance for each planned travel;as a result,the images taken by the Yutu 2 rover differ saliently in scale,rotation,and translation.Consequently,the overlap areas of these images have small sizes,especially when the resolution and morphology of the overlap areas are very different,thereby resulting in image matching difficulties to image matching,and hindering automatic visual positioning between adjacent stations.In this paper,a general pixel-spatial resolution model focusing on lunar images is proposed by introducing the inclination angle to existing models.We further discuss in detail the variation of image scales and the rate of feature matching in terms of station distances,imaging distances,and lunar surface inclination angles or their combinations,and develop the mechanism to choose the equivalent minimal factor subset.Finally,the proposed model is evaluated by using the station data of the Yutu 2 rover collected on the first and second lunar days.Our analysis classified the influence of the pixel-spatial resolution ratio on the rate of image matching into four types.Thus,a general rule was given for describing the image matching rate under different stations,and the constraints to determine the distance and location of adjacent stations in actual tasks were presented.The proposed approach exhibits the potential to be utilized for improving the automation of localization in subsequent lunar explorations.