毕业生知识库

随着科技的发展和进步，移动机器人逐渐在人类的工业生产和日常生活中扮演了越来越重要的角色。根据环境的特点选择合适的地图构建方式并实现自主定位是移动机器人在大范围环境中进行可靠导航的基础，而使用主动视觉系统可以使机器人对周边环境的观测更加便利，具有重要的理论研究意义和广泛的应用前景。本文面向大范围环境地图构建与导航问题，围绕主动视觉测量与环境地图构建的关键技术，对联动偏转式主动视觉系统的标定及不同模态时的三维测量、基于点线特征的单目相机同时定位与地图构建、基于图像HOG特征的室外环境多层次地图构建开展了一系列研究。本文主要的工作和贡献有：
（1）为了提高立体视觉系统测量的便利性，设计了一种联动偏转式主动视觉系统，并提出了系统标定方法和三维测量方法。该主动视觉系统的两台相机可以同时相向或反向偏转，根据偏转角度的不同分为前视、注视、侧视模态。针对该主动视觉系统，建立了一种具有五个恒定系统参数和五个变量的新型测量模型，并设计了一种切实可行的标定方法对视觉系统的各个参数进行标定。实验表明，所提出的方法可以高精度地测量公共视场中任何特征点的三维坐标，五个常量系统参数的高精度标定有利于提高联动偏转式主动视觉系统的测量精度，并避免对不同偏转角的重复标定。
（2）单目视觉系统难以获取物体的深度信息，通常难以进行三维测量。结合移动平台和联动偏转式主动视觉系统，提出了一种基于运动的单目视觉系统三维测量方法。根据视觉系统随着移动平台平移时的运动增量和目标点的图像坐标变化量，计算出目标点的深度。目标点在笛卡尔空间中的X和Y坐标，用深度和相应的归一化平面内的成像坐标计算得出。在相机正视和相机侧视两个典型例子中，分析了不同变量引起的误差。实验结果验证了所提出的带有偏转相机的单目视觉系统测量方法的准确性和有效性。
（3）针对基于点特征的视觉SLAM在低纹理、光照变化明显场景下的鲁棒性下降问题，提出了一种结合点特征与线特征的视觉SLAM系统。在ORB-SLAM2的基础上加入线特征，使用Plücker坐标对空间直线进行坐标变换和投影，使用正交表示对直线特征进行优化。对空间直线三角测量时的相机退化运动进行了分析，并且设计了一种基于消失点的残差项，使得Plücker坐标中的方向向量可以得到充分校正，减少退化运动对建图精度的影响。实验证明，所提出的方法增强了系统的鲁棒性，提高了系统的位姿估计精度和3D建图精度。
（4）为了充分利用环境的先验知识，使移动机器人在大规模室外环境中实现自主导航和定位，提出了一种根据环境先验知识离线建立的多层次地图。它由拓扑图、全局度量简图、语义地图和局部度量地图组成，可以使用提出的联动偏转式主动视觉系统离线构建。提出了基于节点附近的物体共视关系的节点识别方法，和基于线段的道路感知和基于消失点的导航控制方法。利用不同层次的地图可以实现路径规划、节点识别和相对位姿估计等功能，使移动机器人在大规模室外环境中实现自主导航和定位。在室外环境实验中，机器人利用上述方法实现了导航并运动至目的地，验证了方法的有效性。

With the development and progress of science and technology, mobile robots gradually play a more important role in industrial production and daily life. Choosing an appropriate map model according to the characteristics of the environment and realizing autonomous positioning are the basis for a mobile robot to conduct reliable navigation in a large-scale environment, and it is more convenient for a robot to observe and measure the surrounding environment through an active vision system, which is significant in both theoretical research and real-world applications. This thesis focuses on map construction and navigation of mobile robots with large-scale environments, and studies the key techniques of active vision measurement and map building of environments. A series of studies have been carried out on the calibration and 3D measurement method of active vision system with symmetric yawing cameras, simultaneous localization and mapping of monocular cameras with points and lines, and hierarchical map for the outdoor environments based on HOG features. The main works and contributions of this thesis are summarized as follows:
(1) In order to improve the convenience of 3D measurement with stereo vision system, an active vision system with symmetric yawing cameras is proposed, and the calibration and 3D measurement method of the active vision system is presented. The two cameras are yawed with equivalent angles in opposite directions from their common centerline. and can be divided into forward-looking mode, gazing mode, and side-looking mode. A new measurement model with five constant systematic parameters and five variables is established for the active vision system. Also, a design for a relatively simple and convenient calibration method to determine the constant systematic parameters for the active vision system is presented. The experimental results verify the effectiveness of the proposed measurement model and show that the proposed method can measure the 3-D positions of any feature points in the common FOV with high accuracy. The results demonstrate that these five constant systematic parameters are highly beneficial for active vision systems with symmetric yawing cameras, enabling improved measurement accuracy, and avoidance of repeated calibration for varying yaw angles.

(2) It is difficult for a monocular vision system to obtain the depth information of objects, which results in the failure of 3D measurement. A motion-based measurement method is presented for a monocular vision system consisting of a yawing camera, based on the proposed active vision system with symmetric yawing cameras and the moving platform. The depth of the object is estimated with the motion increment and the image features’ variation when the camera translates with the moving platform. Then the object's X and Y coordinates are computed with the depth and its imaging coordinates. In the two typical cases of the front view and the side view, the errors caused by different variables are analyzed. The experimental results verify the accuracy and effectiveness of the proposed measurement method.

(3) A visual SLAM method based on point and line features is proposed to solve the low robustness of the visual SLAM method based on point features in the environment with low texture and changing illumination. The proposed VSLAM method is on the basis of ORB-SLAM2 and line features are combined with it. Plücker coordinates are used for spatial line’s coordinate transformation and line feature projection, and orthonormal representation is used for optimization. The degenerate motions of the camera during line triangulation are analyzed using Plücker coordinates, and a novel residual term using vanishing points is proposed for optimization to make the line’s direction vector of Plücker coordinates corrected sufficiently and reduce the influence of the degeneracy problem on mapping accuracy. Experimental results show that the proposed method enhances the robustness of the system, and improves the accuracy of pose estimation and 3D map.

(4) In order to make full use of the pre-known knowledge of the environment and to enable mobile robots to realize autonomous navigation and localization in large-scale outdoor environments, a novel hierarchical outdoor map constructed offline based on the pre-known knowledge of the environment is proposed. It consists of a topological map, a simple global metric map, a semantic map, and a local metric map, which can be constructed offline using the proposed active vision system with symmetric yawing cameras. A node recognition method based on the common-viewing relationship of the node’s surrounding objects, a line-segment-based road perception method, and a vanishing point-based navigation control method are proposed. Using the different part of the map can realize path planning, node recognition, and relative pose estimation, enabling mobile robots to achieve autonomous navigation and positioning in large-scale outdoor environments. In the outdoor environment experiment, the robot uses the above method to achieve navigation and move to the destination, which verifies the effectiveness of the method.

Keywords: Mobile robot, Active vision, 3D measurement, VSLAM, Hierarchical map