基于视觉的多机器人协作SLAM研究

发布时间：2018-01-02 01:08

本文关键词：基于视觉的多机器人协作SLAM研究　出处：《哈尔滨工业大学》2016年博士论文　论文类型：学位论文

【摘要】：移动机器人是机器人学的一个重要分支,可应用于未知环境探索、巡逻、服务等诸多领域。目前移动机器人技术还不成熟,多项关键技术需要改进,其中作为实现自主、智能移动机器人前提的即时定位与地图构建(SLAM,Simultaneous Localization and Mapping)技术尤其需要进一步研究。本课题源于国家自然科学基金资助项目“基于局部不变映射的双目移动机器人协作SLAM研究”,深入研究了基于视觉的多机器人协作SLAM及其相关技术。主要内容包括以下几个方面:首先研究了多机器人系统及其任务分配算法。提出了一种基于UPn P(Universal Plug and Play)技术的多机器人系统UMRS(UPn P-based Multi-robot System)。该系统使得成员间具有互发现能力,避免了多机器人系统通常存在的单点故障、协作协议与底层通信耦合度高等问题。在此基础上对多机器人任务分配技术进行了研究,提出一种适用于MT-SR-TA类型任务分配问题的方法CMRTA(CHNN-based Multi-robot Task Allocation)。其次,研究了基于双目视觉的自然路标提取与描述方法。针对SLAM过程中存在的作为路标的特征点过多而导致数据关联复杂度高、准确度低的问题,提出一种以特征点的三维信息为基础的路标提取方法。该方法从基于双目视觉获得的环境图像中提取并匹配特征点,重建特征点对应的空间点的三维信息,并依据点间距离进行聚类分析得到若干点簇,将每个点簇整体作为一个路标。为了便于进行路标间的快速匹配,达到数据关联的目的,对路标进行标识,提出了一种路标描述符,论述了其生成方法和匹配过程。为了获得合适的路标,本文对Mean Shift聚类算法进行了改进,通过最小点数、聚类半径初始值、半径增长幅度、最大聚类半径等参数的调节,使得算法可以根据空间点具体的分布情况产生适当数量的不同尺寸的点簇。再次,对基于视觉的SLAM进行了研究。针对基于EKF(Extended Kalman Filter)的SLAM算法因为计算复杂度过高不适合大规模环境地图构建的问题,提出了一种基于自然路标和局部地图更新的NL-SLAM(Natural landmark and Local map based SLAM)算法。由于自然路标的使用,减少了位姿和地图估计的误差,同时因为路标数量的减少和局部地图的使用,有效降低了计算复杂度。最后,在以上工作的基础上,进一步研究了基于视觉的多机器人协作SLAM。提出一种团队共享路标信息的MR-v SLAM(Multi-robot visual SLAM)算法,该算法改进了Fast SLAM使之适用于多机器人协作SLAM。在该算法中,多机器人系统的每个成员均进行SLAM,将其他成员视为自身传感器的延伸,在SLAM过程中不断将其他成员的观测到的路标信息融合到自己的地图中,加快了对大规模未知环境的地图构建速度。
[Abstract]:Mobile robot is an important branch of robotics, which can be applied to unknown environment exploration, patrol, service and many other fields. At present, mobile robot technology is not mature, many key technologies need to be improved. As the premise of realizing autonomous and intelligent mobile robot, the real-time location and map construction of SLAM are discussed. Simultaneous Localization and Mapping). In particular, further research is needed. This project is a project supported by the National Natural Science Foundation of China, "the study of Binocular Mobile Robot Cooperative SLAM based on Local invariant Mapping". In this paper, the vision-based multi-robot cooperative SLAM and its related technologies are studied in depth. The main contents include the following aspects:. Firstly, the multi-robot system and its task assignment algorithm are studied. A new algorithm based on UPn P(. UMRS(UPn P-based Multi-robot system based on Universal Plug and). The system enables members to discover each other. The problems such as single point failure and high coupling degree between the cooperation protocol and the underlying communication are avoided in the multi-robot system. On this basis, the multi-robot task allocation technology is studied. This paper presents a method for MT-SR-TA type task assignment problem, CMRTA(. CHNN-based Multi-robot Task allocation. The method of extracting and describing natural landmarks based on binocular vision is studied. Aiming at the problems of high complexity and low accuracy of data association caused by too many feature points as landmarks in the process of SLAM. A road sign extraction method based on 3D information of feature points is proposed, which extracts and matches feature points from environmental images obtained from binocular vision, and reconstructs 3D information of spatial points corresponding to feature points. Cluster analysis based on the distance between points to obtain a number of points, each cluster as a whole as a road sign. In order to facilitate the rapid matching between the road signs, to achieve the purpose of data association, the road signs are identified. In this paper, a signpost descriptor is proposed, and its generating method and matching process are discussed. In order to obtain the appropriate signpost, the Mean Shift clustering algorithm is improved, and the minimum number of points is obtained. The adjustment of the initial value of the clustering radius, the radius increase amplitude and the maximum clustering radius makes the algorithm produce a proper number of different size clusters according to the specific distribution of the spatial points. In this paper, SLAM based on vision is studied. Because of the high computational complexity, the SLAM algorithm is not suitable for large-scale environmental map construction. A NL-SLAM(Natural landmark and Local map based slam based on natural road sign and local map updating is proposed. Algorithm. Due to the use of natural road signs. The errors of pose and map estimation are reduced, and the computational complexity is effectively reduced because of the reduction of the number of road signs and the use of local maps. Finally, based on the above work. Furthermore, the vision-based multi-robot cooperative slam is studied. A MR-v SLAM(Multi-robot visual slam is proposed to share the roadmap information in a team. Algorithm. The algorithm improves Fast SLAM and makes it suitable for multi-robot cooperative slam. In this algorithm, every member of a multi-robot system performs SLAM. The other members are regarded as the extension of their own sensors. In the process of SLAM, the information of the road signs observed by other members is continuously fused into their own maps, which speeds up the construction of maps of large-scale unknown environments.
【学位授予单位】：哈尔滨工业大学
【学位级别】：博士
【学位授予年份】：2016
【分类号】：TP242

【参考文献】