【作者】 刘康；

【导师】 郭磊；

【作者基本信息】 北京邮电大学 ， 控制科学与工程， 2018， 硕士

【摘要】 近年来,随着图像处理技术、计算机技术的快速发展,视觉技术被广泛的应用于机器人领域中。基于视觉的机器人能够自主的获取周围环境信息,可以用来进行机器人的自主定位、姿态测量、路径规划、导航、控制等。在机器人研究过程中,通常会对机器人的结构、控制系统、机器人的环境感知、机器人的定位和姿态测量等进行研究。本次研究的课题是“基于视觉的机器人姿态测量”。主要研究内容为基于视觉机器人的姿态测量,可以分为如下两部分:理论部分:实现了一种基于视觉的姿态测量方法,利用固定的合作矩形目标,将摄像头采集的图像进行灰度化、二值化、中值滤波去噪等预处理,然后对预处理后的图像利用ORB算法提取矩形区域的特征点,对提取到的特征点用K-Means算法进行聚类,得到矩形的四个角点作为姿态解算的特征点,最后用PNP算法对姿态进行解算,得到相机在世界坐标系下的姿态信息。同时针对此算法设计了一种实验平台,实验平台可以实现绕一个轴在平面内进进行旋转,同时在实验平台上搭载了姿态传感器MPU6050。当旋转实验平台时,对比MPU6050和视觉角度的变化量误差稳定在2.5度左右,验证了算法的有效性。最后分析了本文实验中的主要误差来源,针对误差来源提出了一种基于亚像素的角点提取方法,在相同实验条件下,此方法能够提高姿态的测量精度。实验部分:在第一部分的基础上设计了一种基于视觉姿态测量的三轴转动机器人平台,此机器人平台可以带动相机实现笛卡尔坐标系内分别沿x轴、y轴、z轴的独立转动。其中主要设计了机器人平台的机械结构部分包括关键构件的设计和选型,同时还设计了机器人平台的控制电路部分。最后针对视觉的姿态测量,提出了利用卡尔曼滤波的方法将视觉姿态和惯性姿态相融合,并在Matlab中进行了仿真,仿真结果表明卡尔曼滤波器对于噪声信息有着有效的滤除作用。更多还原

【Abstract】 In recent years,with the rapid development of image processing technology and computer technology,computer vision technology has been widely used in the field of robotics.Robot based on vision can acquire information of surrounding environment autonomously,and it can be used for robot’s autonomous positioning,attitude measurement,path planning,navigation and control.In the process of robot research,we usually study the robot’s structure,control system,robot’s environment perception,robot’s position and attitude measurement.Attitude measurement of robot based on vision is the subject of this study.The main research content is attitude measurement based on visual robot,which can be divided into two parts as follows:The theoretical part:A kind of attitude measurement method based on vision is realized by using cooperation of the rectangular target fixed on the blackboard.The image acquired by the camera,after a series of preprocess of grayscale,binarization,median filtering denoising,and then using ORB algorithm to extract the feature points of the rectangular region.The four corner points of the rectangular can be extracted by the K-Means clustering algorithm,and then the camera attitude in the world coordinate system can be solved by PNP algorithm.Finally,an experimental platform is designed for the algorithm.The experimental platform can rotate around a shaft in the plane.At the same time,the attitude sensor MPU6050 is mounted on the experimental platform.When rotating the experimental platform,the effectiveness of the algorithm is verified by comparing the change of MPU6050 and visual angle.The change of MPU6050 and visual angle.Finally,the main error sources in the experiment are analyzed,and a corner extraction method based on sub-pixel is proposed to eliminate error.The experimental shows that this method can improve the accuracy of attitude measurement.The experimental part:On the basis of the first part,a three axis rotation robot platform based on visual attitude measurement is designed,this robot platform can drive the independent rotation of the camera along the X axis,the Y axis and the Z axis respectively in the Cartesian coordinate system.The mechanical part of the robot platform is designed,including the design and selection of the key components,and then the control circuit part of the platform is also designed.Then,for the designed three axis rotating robot platform,the dynamic model of the camera’s rotation process is built.Finally,this paper proposes a method to fuse the attitude acquired by vison and attitude acquired by inertial unit using Kalman filter.Besides,simulation verification is carried out in MATLAB,the simulation results show that the Kalman filter has an effective filtering effect on the noise information.更多还原