【作者】 冷春辉；

【导师】 李惠光；

【作者基本信息】 燕山大学 ， 模式识别与智能系统， 2010， 硕士

【摘要】 机器人视觉伺服是机器人研究领域所涉学科众多、最富有挑战性的课题之一,其综合了控制理论、机器人学、计算机视觉、数字图像采集处理、嵌入式系统、网络通信以及科学计算等不同领域的知识。目前,机器人视觉伺服主要分为基于位置的视觉伺服和基于图像的视觉伺服两个较大的研究类别。基于位置的视觉伺服根据已知的像机内、外参数对目标进行三维建模,对摄像机标定和建模精度要求比较高;与之相比基于图像的视觉伺服则对像机、机器人和环境建模误差具有较强的鲁棒性,但多数文献都桎梏于如何解决深度的估计问题。本文在系统总结了目前机器人视觉伺服研究领域发展状况的基础上,主要针对双目视觉系统研究了以下几个方面内容:首先,使用基于C/C++语言的计算机视觉开放函数库OpenCV完成双目视觉系统中左右眼摄像机的单独标定,进而给出两摄像外参数与它们的相互位置关系即结构参数的变换关系,完成双目视觉系统标定工作。其次,在实现特征点匹配任务中,将RANSAC算法应用于SIFT特征匹配,进而提出了一种基于RANSAC算法的双向纠错SIFT特征匹配改进算法;给出了一种基于角点检测的特征点匹配算法,该方法成功应用于视觉伺服控制实验中的目标特征点实时匹配。再次,将双目视觉伺服控制模型应用于实际的机器人视觉伺服控制系统中,在以MOTOMAN-UP6机械手为被控对象的双目视觉伺服系统硬件平台上,采用VC6.0软件编程,通过调用机械手控制器串行通信接口函数控制机械手运动,编制了图像采集与处理程序、机器人运动与控制程序、控制算法程序,搭建了双目视觉伺服控制的实验研究平台。最后,针对实验过程中出现的机器人末端平移运动与旋转运动相互耦合的问题,提出了一种开关控制方法。该开关控制方法借鉴滑模变结构的控制思想,将整个伺服控制过程分为调姿—定位—精确定位三个阶段,有效的解决了采用单一结构形式的图像雅可比矩阵进行伺服控制时可能出现的耦合问题。最后进行了基于MOTOMAN-UP6机械手的双目视觉伺服控制对比实验,结果表明:该开关控制方法具有更好的稳定性和定位轨迹,较好的解决了在实际控制过程中机器人末端出现的平移运动与旋转运动相互耦合问题。更多还原

【Abstract】 Robot visual servoing is one of the most active and challenging topics in robotics, and is also involved in many diverse research fields, such as control theory, robotics, computer vision, digital image acquisition and processing, embedded systems, network communications, scientific computing and so on. Currently speaking, the research of visual servoing could be divided into two major categories: Position-Based Visual Servoing(PBVS) and Image-Based Visual Servoing(IBVS). PBVS calculates the relative position and orientation, which are known as the 3D modeling of the object according to the known camera internal and external parameters. However, it needs perfect target geometric model and largely depends on accurate camera calibration. Comparing with PBVS, the IBVS is object model free, and robust to camera modeling and hand-eye calibration errors.In this paper, on the basis of giving a tutorial introduction to the development of robot visual servoing control system nowadays, the main work is summarized as follows:First of all, a camera calibration method of binocular visual system is proposed in this paper, the method finishes the two separate camera calibration using the function of the Open Source Computer Vision Library at first, and then calculates the conversion relationship between the relative pose of the two cameras using the camera external parameters.Secondly, in the feature points matching tasks, a bidirectional matching algorithm based RANSAC algorithm is proposed in this paper to improve the image matching accuracy of the SIFT algorithm, and also a matching algorithm based on corner detection function of OpenCV is introduced, in which we extracts the feature points of the target object using strong Harris corner detection algorithm and matching them using location constraint. Thirdly, Based on the hardware and operation environment of the robot MOTOMAN-UP6, an experiment operation platform is established and the programs of image sampling and processing, robot motion and control, controller algorithm are developed. An experiment is carried to verify the visual servoing model irrespective of depth. The analysis of the error shows that the model is validity and practicability.Finally, a switching control method is proposed to decouple the rotational and translational motion control of the robot end-effector and avoid the inherent drawbacks of the origin model in practical application. The whole switching control method consists of three steps: adjusting posture; positioning; precision positioning (posture and position). The method is implemented and validated on a MOTOMAN-UP6 based eye-in-hand platform and the experimental comparison shows that it can enhance the system stability and have a better positioning trajectory.更多还原