双目立体视觉SLAM特征匹配与定位技术研究

发布时间：2018-06-26 14:48

本文选题：双目立体视觉 + 同时定位与地图构建　；参考：《东华大学》2017年硕士论文

【摘要】：同步定位与地图构建(Simultaneous location and mapping,SLAM)是指机器人在未知环境中创建环境地图并推断自身位姿的过程。近年来,SLAM问题的解决对于移动机器人实现自主定位具有十分重要的意义,已逐渐成为移动机器人导航及计算机视觉领域的热门研究课题。由于SLAM中机器人的特征点匹配往往出现误匹配且匹配的复杂度较高而导致机器人构建地图的周期较长,定位的实时性较差。为此,我们对SLAM的特征匹配与定位技术展开理论研究。本文针对双目立体视觉SLAM系统展开研究,具体工作如下:(1)介绍了SLAM系统框架,分析了两种常用的SLAM滤波算法—扩展卡尔曼滤波器算法和粒子滤波器算法。给出了本文研究双目立体视觉SLAM的观测与运动两大基本模型,在此基础上提出了本文研究的双目立体视觉SLAM的整体结构框架。(2)对双目立体视觉SLAM的数据关联问题进行了研究。介绍了特征提取与匹配技术的相关知识,并针对SIFT特征提取与匹配算法的维数较大且存在较大的误匹配率问题,结合支持向量机(SVM)的序列最小优化算法(SMO)进一步细匹配提出基于序列最小优化的SIFT特征提取与匹配算法—SMO-SIFT算法。最后通过MATLAB仿真表明SMO-SIFT算法降低了算法的维数,改善了算法的实时性,同时提高了算法精确度。(3)对SLAM的路径估计问题进行了研究。介绍了RaoBlackwellised粒子滤波器算法(RBPF)并针对RBPF算法的粒子数目的增加会频繁重采样从而导致“粒子退化”问题,提出了基于小生境遗传优化算法的INGO-RBPF算法。最后通过MATLAB仿真表明INGO-RBPF算法具较高的估计精度和稳定性,抗干扰能力较强,比较适合应用在SLAM实时定位中。(4)在机器人操作系统(ROS)的环境下将SMO-SIFT和INGORBPF算法运用于实验环境中。给出了SLAM系统中的地图构建、机器人控制及远程控制三大模块的软硬件设计,实验结果表明机器人能够正确的构建出环境地图和成功定位,运行结果比较理想。
[Abstract]:Synchronous location and mapping (slam) is a process in which robots create environmental maps and infer their posture in unknown environments. In recent years, the solution of slam problem is very important for mobile robot to achieve autonomous positioning, and has gradually become a hot research topic in the field of mobile robot navigation and computer vision. Due to the false matching of robot feature points in slam and the high complexity of matching, the robot has a long period of map construction and poor real-time localization. Therefore, the feature matching and localization techniques of slam are studied theoretically. The main work of this paper is as follows: (1) the framework of slam system is introduced, and two commonly used slam filtering algorithms, extended Kalman filter algorithm and particle filter algorithm, are analyzed. In this paper, two basic models of observation and motion of binocular stereo slam are presented. Based on these two models, the whole frame of binocular stereo slam is proposed. (2) the data association problem of binocular stereo slam is studied. This paper introduces the knowledge of feature extraction and matching, and aims at the problem that sift feature extraction and matching algorithm has large dimension and large mismatch rate. Combined with support vector machine (SVM) sequence minimum optimization algorithm (SMO), a SMO-SIFT feature extraction and matching algorithm based on sequential minimum optimization is proposed. Finally, MATLAB simulation shows that SMO-SIFT algorithm reduces the dimension of the algorithm, improves the real-time performance of the algorithm, and improves the accuracy of the algorithm. (3) the path estimation problem of slam is studied. In this paper, Rao Blackwellised particle filter algorithm (RBPF) is introduced. Aiming at the problem of "particle degradation" caused by increasing the number of particles in RBPF algorithm, an INGO-RBPF algorithm based on niche genetic optimization algorithm is proposed. Finally, MATLAB simulation shows that INGO-RBPF algorithm has high estimation accuracy and stability, strong anti-jamming ability, and is more suitable for real-time localization of slam. (4) SMO-SIFT and INGORBPF algorithms are applied in the environment of robot operating system (Ros). The software and hardware design of the three modules in slam system are given. The experimental results show that the robot can correctly construct the map of environment and locate successfully, and the result of running is ideal.
【学位授予单位】：东华大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP242;TN713

【参考文献】