基于RGB-D的室内场景SLAM方法研究
发布时间：2018-08-08 14:34
【摘要】：近年来,随着计算机技术的迅猛发展,同时定位与建图(Simultaneous Localization and Mapping,SLAM)技术在移动机器人、无人机、无人驾驶、视觉医疗、AR/VR、可穿戴设备等方面得到了广泛的应用。随着图优化问题中稀疏矩阵的发现,基于视觉的SLAM方法已经成为国内外的研究热点,基于图优化的SLAM方法逐渐应用于在大规模场景中。本文采用华硕Xtion Pro Live深度相机作为传感器,提出了一种基于改进BoVW模型的三维SLAM方法,在本文提出的SLAM方法中,在图像检测和闭环检测算法上提出了改进,并通过实验证明提高SLAM的效率和鲁棒性。首先,介绍了基于视觉SLAM的基本原理和方法。对SLAM问题进行了描述,分析了几种经典的SLAM方法,对比了几种经典的特征检测的优缺点,针对视觉SLAM对图像特征提取的要求,在ORB特征提取算法上提出了一种基于自适应的区域分割ORB特征提取方法。并在图像特征匹配方法上,采用传统的随机采样一致性(Random Sample Consensus,RANSAC)算法和K近邻(K-Nearest Neighbor algorithm,KNN)算法消除误匹配,有效地减少误匹配点数,提高了匹配的精度和速度。在点云数据融合算法上,采用迭代最近(Iterative Closet Point,ICP)算法,用奇异分解(SVD)方法进行求解计算相机位姿。其次,在闭环检测方法上,介绍了闭环检测的作用和方法,及闭环检测中的问题和难点,在基于BoVW模型的闭环检测方法中,介绍了视觉词典的创建方法,相对于传统K-Means聚类算法的缺点,提出了一种改进的K-Means算法,有效地解决了K-Means算法依赖初始聚类中心,容易陷入局部最优的问题,提高了闭环检测的准确率。最后,设计了一种基于RGB-D的室内场景SLAM系统,并通过实验把本文改进的算法应用于该SLAM方法中。
[Abstract]:In recent years, with the rapid development of computer technology, simultaneous location and mapping (Simultaneous Localization and mapping slam) technology has been widely used in mobile robots, unmanned aerial vehicles, visual medical AR-VR, wearable devices and so on. With the discovery of sparse matrix in graph optimization problem, SLAM method based on vision has become a hot topic at home and abroad, and SLAM method based on graph optimization is gradually applied in large-scale scene. In this paper, using Asus Xtion Pro Live depth camera as sensor, a 3D SLAM method based on improved BoVW model is proposed. In the proposed SLAM method, the image detection and close-loop detection algorithms are improved. The experimental results show that the efficiency and robustness of SLAM are improved. Firstly, the basic principle and method of visual SLAM are introduced. This paper describes the SLAM problem, analyzes several classical SLAM methods, compares the advantages and disadvantages of several classical feature detection methods, and aims at the requirements of visual SLAM for image feature extraction. An adaptive region segmentation ORB feature extraction method based on ORB feature extraction algorithm is proposed. In the image feature matching method, the traditional random sampling consistent (Random Sample ConsensusRANSAC algorithm and the K-Nearest Neighbor algorithm KNN algorithm are used to eliminate the mismatch, which can effectively reduce the number of mismatch points and improve the accuracy and speed of the matching. In the point cloud data fusion algorithm, the iterative nearest (Iterative Closet Point ICP algorithm and singular decomposition (SVD) method are used to calculate the camera pose. Secondly, the function and method of closed-loop detection are introduced, and the problems and difficulties in closed-loop detection are introduced. In the close-loop detection method based on BoVW model, the method of creating visual dictionary is introduced. Compared with the traditional K-Means clustering algorithm, an improved K-Means algorithm is proposed, which effectively solves the problem that the K-Means algorithm depends on the initial clustering center and is prone to fall into the local optimal condition, and improves the accuracy of closed-loop detection. Finally, an indoor scene SLAM system based on RGB-D is designed, and the improved algorithm is applied to the SLAM method through experiments.
【学位授予单位】：湖南工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
本文编号：2172093
[Abstract]:In recent years, with the rapid development of computer technology, simultaneous location and mapping (Simultaneous Localization and mapping slam) technology has been widely used in mobile robots, unmanned aerial vehicles, visual medical AR-VR, wearable devices and so on. With the discovery of sparse matrix in graph optimization problem, SLAM method based on vision has become a hot topic at home and abroad, and SLAM method based on graph optimization is gradually applied in large-scale scene. In this paper, using Asus Xtion Pro Live depth camera as sensor, a 3D SLAM method based on improved BoVW model is proposed. In the proposed SLAM method, the image detection and close-loop detection algorithms are improved. The experimental results show that the efficiency and robustness of SLAM are improved. Firstly, the basic principle and method of visual SLAM are introduced. This paper describes the SLAM problem, analyzes several classical SLAM methods, compares the advantages and disadvantages of several classical feature detection methods, and aims at the requirements of visual SLAM for image feature extraction. An adaptive region segmentation ORB feature extraction method based on ORB feature extraction algorithm is proposed. In the image feature matching method, the traditional random sampling consistent (Random Sample ConsensusRANSAC algorithm and the K-Nearest Neighbor algorithm KNN algorithm are used to eliminate the mismatch, which can effectively reduce the number of mismatch points and improve the accuracy and speed of the matching. In the point cloud data fusion algorithm, the iterative nearest (Iterative Closet Point ICP algorithm and singular decomposition (SVD) method are used to calculate the camera pose. Secondly, the function and method of closed-loop detection are introduced, and the problems and difficulties in closed-loop detection are introduced. In the close-loop detection method based on BoVW model, the method of creating visual dictionary is introduced. Compared with the traditional K-Means clustering algorithm, an improved K-Means algorithm is proposed, which effectively solves the problem that the K-Means algorithm depends on the initial clustering center and is prone to fall into the local optimal condition, and improves the accuracy of closed-loop detection. Finally, an indoor scene SLAM system based on RGB-D is designed, and the improved algorithm is applied to the SLAM method through experiments.
【学位授予单位】：湖南工业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41
【参考文献】
相关期刊论文 前5条
1 王忠立;赵杰;蔡鹤皋;;大规模环境下基于图优化SLAM的后端优化方法[J];哈尔滨工业大学学报;2015年07期
2 陈白帆;蔡自兴;邹智荣;;一种移动机器人SLAM中的多假设数据关联方法[J];中南大学学报(自然科学版);2012年02期
3 朱代先;王晓华;;基于改进SIFT算法的双目视觉SLAM研究[J];计算机工程与应用;2011年14期
4 温丰;柴晓杰;朱智平;董小明;邹伟;原魁;;基于单目视觉的SLAM算法研究[J];系统科学与数学;2010年06期
5 陈伟;吴涛;李政;贺汉根;;基于粒子滤波的单目视觉SLAM算法[J];机器人;2008年03期
相关硕士学位论文 前3条
1 郑顺凯;自然环境中基于图优化的单目视觉SLAM的研究[D];北京交通大学;2016年
2 刘芳;动态未知环境下移动机器人同时定位与地图创建[D];哈尔滨工业大学;2015年
3 陈超;基于TOF摄相机的三维点云地图构建研究[D];哈尔滨工业大学;2013年
，本文编号：2172093
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2172093.html
