基于双目视觉的移动机器人室内三维地图构建方法研究

发布时间：2018-06-06 05:27

本文选题：移动机器人 + 三维地图构建　；参考：《哈尔滨工业大学》2017年硕士论文

【摘要】：随着移动机器人在工业制造、未知环境探索、人类服务、军事等领域的广泛应用,研究如何实现移动机器人的自主定位与导航一直是机器人领域的热点问题。1988年提出的SLAM(Simultaneous Localization and Mapping,即时定位与构图)技术被众多研究人员认为是实现移动机器人自主化的关键。而在SLAM技术中地图构建作为其中重要的组成部分,其精度和完整性是SLAM技术得以实现的基础。目前,在移动机器人地图构建理论和方法的研究中,三维地图相对于二维地图实用价值更高,应用更加广泛,且研究成果少,对于研究出环境适应性强,高效实用的三维地图构建方法有着迫切的需要。本文的研究目的是针对移动机器人三维地图构建中的相关问题,对移动机器人三维地图构建中的关键技术展开研究,并在室内环境下验证相关算法的可行性。围绕这一目的,主要进行了以下的研究工作。设计并搭建了基于双目视觉的移动机器人三维地图构建平台。本文深入分析了移动机器人运行环境特点,采用系统集成的方法完成室内环境下移动机器人三维地图构建平台的设计与调试。在充分了解平台的运动特性和机械结构基础上,建立了机器人平台运动模型,并采用航位推算方法实现了机器人相对位姿估计。研究基于立体视觉的三维点云数据获取方法。文中在对摄像机几何光学模型和畸变模型进行分析的基础上,对双目视觉中涉及的立体匹配等算法展开研究。着重探讨了影响立体匹配效果的因素,并结合三维地图构建的实际要求,通过边缘检测的方法获取环境中关键边缘特征,并通过双目测距原理完成空间中单帧三维关键特征点云数据获取。研究点云预处理和三维点云配准方法。文中针对室内环境下的粗糙地面特征及光滑地面反射特征问题,提出了基于最小二乘法的地面特征去除方法,并结合点云过滤器和统计分析方法完成三维点云配准前的预处理。根据点云分布特征及位姿估计精度,选择正态分布算法完成连续多帧点云图的拼接及配准,并提出了基于边缘特征的三维点云配准方法。本文比较了边缘关键特征点云和普通稀疏点云在点云配准中的实际效果,验证了本文提出的点云配准方法的有效性。为了验证本文所研究的三维地图构建方法的可行性,文中选择典型的室内楼道环境,控制系统中的移动机器人平台,完成室内环境下全局三维地图的创建,并对三维地图构建结果进行了效果分析和关键影响因素分析。实验结果表明本文所构建的移动机器人三维地图能够满足移动机器人自主定位和导航的要求。
[Abstract]:With the wide application of mobile robots in industrial manufacturing, unknown environment exploration, human service, military and other fields, The research on how to realize autonomous localization and navigation of mobile robots has always been a hot issue in the field of robot. In 1988, the technology of slam localization and mapping was considered by many researchers to be the key to realize autonomous mobile robot. Map construction is an important part of slam technology, and its precision and integrity are the foundation of slam technology. At present, in the research of mobile robot map construction theory and method, 3D map has higher practical value, wider application and less research results compared with two-dimensional map, so it has strong adaptability to research environment. Efficient and practical three-dimensional map construction method has an urgent need. The purpose of this paper is to study the key technology of 3D map construction of mobile robot and verify the feasibility of the algorithm in indoor environment. Around this purpose, mainly carried out the following research work. A three-dimensional map building platform for mobile robot based on binocular vision is designed and built. In this paper, the characteristics of mobile robot running environment are deeply analyzed, and the design and debugging of mobile robot 3D map construction platform in indoor environment are completed by system integration method. On the basis of fully understanding the motion characteristics and mechanical structure of the platform, the motion model of the robot platform is established, and the relative position and attitude estimation of the robot is realized by using the method of dead-reckoning. Three-dimensional point cloud data acquisition method based on stereo vision is studied. Based on the analysis of geometric optical model and distortion model of camera, the stereo matching algorithms involved in binocular vision are studied in this paper. This paper mainly discusses the factors that affect the effect of stereo matching, and combines with the actual requirements of 3D map construction, obtains the key edge features of the environment by edge detection. The key feature cloud data in a single frame is obtained by binocular ranging principle. Point cloud preprocessing and three-dimensional point cloud registration are studied. In order to solve the problem of rough ground feature and smooth ground reflection feature in indoor environment, a method of ground feature removal based on least square method is proposed, and the pre-processing of 3D point cloud registration is completed by combining point cloud filter and statistical analysis method. According to the distribution characteristics of point clouds and the accuracy of position and pose estimation, the normal distribution algorithm is selected to complete the stitching and registration of continuous multi-frame point cloud images, and a 3D point cloud registration method based on edge features is proposed. In this paper, the effectiveness of the point cloud registration method proposed in this paper is verified by comparing the effect of point cloud registration with that of common sparse point cloud. In order to verify the feasibility of the 3D map construction method studied in this paper, the paper selects the typical indoor corridor environment and the mobile robot platform in the control system to complete the creation of the global 3D map in the indoor environment. The result of 3D map construction is analyzed and the key factors are analyzed. The experimental results show that the 3D map of mobile robot can meet the requirements of autonomous localization and navigation of mobile robot.
【学位授予单位】：哈尔滨工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP242

【参考文献】