CORC  > 自动化研究所  > 中国科学院自动化研究所  > 毕业生  > 博士学位论文
题名单目视觉同步定位与地图创建技术研究
作者顾照鹏
学位类别工学博士
答辩日期2011-12-03
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师胡占义 ; 董秋雷
关键词同步定位与地图创建 增强现实 扩展卡尔曼滤波器 惯性传感器 Kinect Monocular SLAM Augmented Reality Extended Kalman Filter Inertial Measurement Unit(IMU) Kinect
其他题名A Study on Monocular Simultaneous Localization and Mapping
学位专业控制理论与控制工程
中文摘要同步定位与地图创建(Simultaneous Localization and Mapping,SLAM)技术指的是移动物体在自身位置不确定的情况下,利用自身的传感器,在未知的环境中创建一个与环境相一致的地图,并同时确定自身在地图上位置的技术。近年来随着计算机视觉技术的进一步发展以及计算机信息处理能力的进一步增强,基于单目视觉的同步定位与地图创建技术(Monocular SLAM)逐渐成为一个研究热点,在机器人定位导航、计算机游戏、增强现实等诸多领域都有着非常广阔的应用前景。本文针对基于单目视觉的同步定位与地图创建技术中存在的一些问题进行了探索和研究,主要工作包括: (1) 提出了一种基于惯性传感器加速度信息的实时单目视觉同步定位与地图创建方法。该方法在扩展卡尔曼滤波器(Extended Kalman Filter,EKF)框架下,将惯性传感器输出的加速度信息实时地引入到滤波器的过程模型中,以有效地降低系统累积误差,抑制系统状态向量的漂移。此外,为了降低数据关联的不确定性,保证滤波器观测模型中的观测向量与地图特征向量之间的正确匹配,在数据关联过程中引入了一种基于角点的分层数据关联算法。 (2) 提出了一种基于惯性传感器横滚角和俯仰角的实时单目视觉同步定位与地图创建方法。针对中低档惯性传感器输出的偏航角误差较大的情况,该方法首先利用惯性传感器输出的横滚角和俯仰角进行系统标定;然后将惯性传感器自身的偏航角作为系统状态向量的一个分量,利用扩展卡尔曼滤波器实时地估计状态向量,进而实现实时鲁棒的同步定位和地图创建。实验结果表明,该方法可以有效地抑制SLAM系统运行过程中产生的累积误差,并降低惯性传感器测量误差对SLAM系统稳定性的影响. (3) 提出了一种基于分层二分图模型的实时重定位算法。当SLAM系统定位失败时,该算法利用分层二分图模型计算当前帧的图像特征与地图特征之间的对应关系,进而实时地实现SLAM系统的重定位。该算法中构建的二分图模型综合利用了图像特征与地图特征的时间和空间约束,有效地提高了系统重定位的鲁棒性。 (4) 实现了一种基于编码模板的高精度实时头部混合定位系统VisTracker-barcode。该系统通过对摄像机采集到的视频图像进行模板识别,获得测量信息,利用扩展卡尔曼滤波器融合惯性传感器输出的运动信息,获得高精度的摄像机位置和姿态。此外,针对实际系统中摄像机与惯性传感器的数据同步采集问题,提出了一种利用多状态扩展卡尔曼滤波的混合定位方法,有效降低了异步数据对系统定位精度的影响。 (5) 提出了一种基于单目视觉SLAM和Kinect的室内场景重建算法。该算法利用基于扩展卡尔曼滤波的同步定位与地图创建方法实时地计算Kinect彩色摄像机的位置与姿态,进而融合Kinect输出的场景深度信息,获取稠密的场景三维点云。
英文摘要Simultaneous Localization and Mapping (SLAM) is a technique used by mobile objects with sensors to build up a consistent map within an unknown scene, while tracking their current locations at the same time. In recent years, monocular SLAM has attracted much attention with the advance of both the computer vision technology and the computer's computational capacity, and it has been applied widely in a variety of applications, such as robot navigation, computer game, augmented reality, etc. This thesis is focused on some key aspects of monocular SLAM, and the main contributions are summarized as: (1) A real-time monocular SLAM method with the acceleration information from an IMU(Inertial Measurement Unit) is proposed within the framework of Extended Kalman Filter(EKF). In this method, the acceleration information outputted from an IMU is introduced to the filter's process model for reducing the accumulated system error as well as the drift of the state vector.Simultaneously, a corner-based hierarchical data association algorithm is introduced to reduce the uncertainty of data association and the likelihood of mismatch between the observation vectors in the filter's observation model and the feature vectors in the map. (2) A monocular SLAM method based only on Roll and Pitch output from an IMU is proposed. We observed that among the three output Euler angles from a low-cost IMU, Roll and Pitch are more accurate than Yaw, then a calibration algorithm for the SLAM system using only Roll and Pitch is presented. With the calibrated results, an IMU-vision SLAM system is implemented robustly, where an EKF is employed to estimate the state vector composed of the camera location, Yaw of the IMU and the locations of the map features. Experimental results show that our proposed method can reduce the accumulated errors effectively and alleviate the negative influence of the IMU’s measurement errors on the system’s stability. (3) A real-time relocalization algorithm based on a hierarchical bipartite graph model is proposed for recovering an SLAM system automatically after tracking failures. When the SLAM system is lost, a hierarchical bipartite graph model is introduced for finding sufficiently good correspondences between the detected image features and the stored map features, and then the efficient real-time relocalization is achieved with the obtained correspondences. The proposed model takes both temporal and spatial constraints on these features into account, ...
语种中文
其他标识符200818014628035
内容类型学位论文
源URL[http://ir.ia.ac.cn/handle/173211/6409]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
顾照鹏. 单目视觉同步定位与地图创建技术研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2011.
个性服务
查看访问统计
相关权益政策
暂无数据
收藏/分享
所有评论 (0)
暂无评论
 

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。


©版权所有 ©2017 CSpace - Powered by CSpace