【作者】 赵斌；

【导师】 刘鹏；

【作者基本信息】 哈尔滨工业大学 ， 计算机科学与技术， 2019， 硕士

【摘要】 深空探测活动已经成为空间科研活动的热点,包括俄罗斯,美国,日本等在内的国家和地区近年来都在积极进行深空探测活动。与此同时,由于深空探测的资金投入大,探测周期长,科技水平高等因素,人们对深空探测追求的回报也不断提高。先前将探测器送达目标天体,传回图像等数据的方式已经无法满足人们的需求。人们希望探测器能够着陆在天体表面,获取岩石土壤样本并带回地球开展研究。而这就要求探测器能够在天体的表面实现软着陆。着陆的第一步就是需要在对目标天体观测的基础上选择着陆点。而着陆点的选择需要获取天体表面的地形,天体的运动规律等数据,从而判断出适合着陆的区域。并且岩石无疑也是一个很重要的衡量点,准确识别出岩石的位置和尺寸对选择着陆点有重要参考价值。本文针对小天体探测活动,目的是自动分析出小天体表面地形条件适合着陆的区域。在获得小天体高分辨率的三维点云模型的前提下,本文对小天体表面地形进行地形特征描述和地形分类,识别出在地形要求上适合着陆的区域。然后对识别出的可能适合着陆的区域进行岩石检测。这些共同为最终的着陆点的选择提供地形,障碍物方面的参考。本文主要研究工作有:(1)首先完成从三维点云模型到高程图的转换,采取拟合椭球体或者拟合平面的思想,计算三维点到基准面的距离作为高程值。保留足够多的细节,为之后的研究工作准备数据。(2)在高程图上划分若干局部区域,利用粗糙度和坡度进行地形特征描述,并且采用卷积神经网络训练分类器对人为划分的几类地形进行地形分类,识别出地形比较平坦的局部区域作为候选着陆区域;(3)在前面工作的基础上,针对识别出的地形平坦的局部区域高程图,采用阈值截取地形凸起处的点,利用聚类算法进行聚类,从而确定岩石的数量和位置,并利用局部区域坡度图调整岩石的范围获得岩石的大致尺寸。最后在高程图上标示出岩石的所在位置和大致范围,完成对候选着陆区域高程图的岩石检测。最后,本文在计算机仿真生成的局部地形和对真实场景重建的三维点云模型数据上进行了实验。实验结果表明本文提出的方法能够有效将三维点云模型变换为高程图,能够准确识别出地形相对平坦的区域,并且可以在平坦区域准确检测出岩石。更多还原

【Abstract】 In recent years,deep space exploration has become a hotspot of space scientific research activities,including Russia,the United States,Japan and other countries and regions are actively engaged in deep space exploration activities.At the same time,due to the large investment of capital,long period of exploration and high level of science and technology in deep space exploration,people are pursuing higher and higher returns.Previously,the way of sending the probe to the target celestial body and returning data such as images has been unable to meet people’s needs.It is hoped that the probe will be able to land on the surface of the celestial body,obtain rock and soil samples and bring them back to Earth for research.This requires that the probe can achieve a soft landing on the surface of the celestial body.The first step of landing is to select the landing site based on the observation of the target celestial body.The choice of landing site needs to acquire the surface terrain of celestial body and the motion law of the celestial body,so as to determine the suitable landing area.And undoubtedly,rock is an important measurement point.Accurate identification of rock’s location and size has great reference value for landing site selection.Aiming at the small body exploration activities,the purpose of this paper is to automatically analyse areas on the surface of small body suitable for landing.On the premise of obtaining the high resolution three-dimensional point cloud model of the small body,terrain characteristic description and terrain classification of the surface terrain of the small body are carried out to identify the areas suitable for landing on the topographic requirements.Then rock detection is performed on the identified areas that may be suitable for landing.These together provide a reference for landing site selection in terms of terrain and obstacles.The main research work of this paper are as follows:(1)Firstly,we need to complete the transformation from three-dimensional point cloud model to elevation map,and take the idea of fitting spheroid or plane to calculate the distance between three-dimensional point and datum plane as elevation value.Ensure enough details to prepare data for later research.(2)Several local areas are divided on the elevation map,and terrain characteristic description is carried out by using roughness and slope.Convolutional neural network training classifier is used to classify the terrain artificially divided into several classes,and the local areas with relatively flat terrain are identified as candidate landing areas.(3)On the basis of the before work,aiming at the identified local elevation map with flat terrain,the threshold is used to intercept the points at the topographic salient,and clustering algorithm is used to determine the number and location of rocks,and the approximate size of rocks is obtained by adjusting the range of rocks with the local slope map.Finally,the location and approximate range of rocks are shown on the elevation map,and the rock detection of the elevation map of the candidate landing area is completed.Finally,experiments are carried out on the local terrain generated by computer simulation and the 3-D point cloud model data reconstructed from the real scene.The experimental results show that the proposed method can effectively transform the three-dimensional point cloud model into elevation map,accurately identify the relatively flat terrain area,and accurately detect rocks in the flat area.更多还原