【作者】 张文昌；

【导师】 黄田；

【作者基本信息】 天津大学 ， 机械工程， 2012， 博士

【摘要】 本文密切结合电子、轻工、食品和医药等领域自动化生产线高速轻载搬运作业需求，研究基于视觉的机器人控制技术、现场快速标定技术，以及机器人-视觉系统与生产线传送带的标定技术，并以开发具备视觉引导功能的少自由度高速包装机器人为工程背景，将研究成果用于提高Delta机器人精度和抓取准确性。全文取得了如下创造性成果：在机器人视觉控制技术方面，构建了基于位置的Dynamic Look-and-Move结构的视觉控制系统方案，为实现机器人对传送带上散乱物料的快速抓放操作提供了重要保障，并在摄像机标定、目标方向定位和多目标动态跟踪抓取三个方面形成特色：(1)摄像机标定。提出一种基于局部线性法的摄像机标定方法，该方法采用通用标定靶，仅需采集一副图片即可确定摄像机的测量精度，进而简化了视觉系统标定过程。(2)目标方向定位。提出一种目标边缘极坐标化和等角度离散的匹配方法，解决了目标绕z轴旋转角度定位问题，具有定位精度高、计算过程简单、实用性强的优点。(3)动态目标跟踪抓取。提出“伺服电机+同步传送带”的动态目标跟踪方法，该方法借助伺服电机编码器实现多目标不重复识别和动态跟踪，并引入牛顿二分法计算动态目标抓取位置，进而实现机器人对动态目标的高频率实时抓取。在机器人运动学标定方面，提出一种基于“单目视觉+激光位移传感器”的机器人现场快速标定方法，标定后机器人精度得到显著提高。该方法在机器人误差建模、误差测量和参数辨识方面形成特色：(1)机器人误差建模。构造了可将影响末端可补偿与不可补偿位姿精度几何误差源有效分离的误差模型，并借助误差灵敏度分析得出边界误差最敏感的结论，在此基础上揭示出仅需检测机器人末端全空间x、y向误差及部分测点z向误差即可有效实现几何误差参数的辨识，为误差测量与参数辨识提供了理论依据。(2)误差测量与参数辨识。提出了一种适用于现场的快速误差测量方法，该方法采用移动摄像机，固定标靶方式，利用基于共面P4P的单目视觉定位原理测量机器人末端在x-y向的定位误差，并借助激光位移传感器测量z向误差，从而有效的辨识出几何误差参数，补偿后机器人定位精度由1mm提高至0.1mm数量级。在机器人-视觉系统与传送带标定方面，提出一种基于单目视觉测量技术的标定方法。该方法通过构造距离度量特征实现视觉系统与传送带之间绕z轴的旋转角度标定，在此基础上借助简易标靶，利用共面P4P的单目视觉测量技术实现机器人与传送带的标定。所提方法为提高机器人系统整体精度提供了重要保障。本文研究成果对丰富和发展视觉机器人技术，推进并联机器人的工程应用具有重要的理论意义和工程实用价值。更多还原

【Abstract】 In close combination with the electron, light industry, food and medicine ofautomation production line for high speed and light load handling requirements, thisdissertation presentsa series of relevant key topics, including robot control based onthe computer vision, fast kinematic calibration, and integrated calibration of visionsystem-robot-production line that can be employed toconfigure a low mobilityhigh-speed parallel robot with vision function.The following contributions have beenmade.A vision-based control system with dynamic position-based look-and-movestructure is designed to perform high frequency pick-and-place tasks on automationproduct line. The major merits can be summarized in brief.(1) Vision system calibration. A camera calibration method based on locallinearization is proposed. By utilizing a square target, only one picture is needed tocalibrate the camera.(2) Target angle recognition. An angle recognition algorithm based on discretearray of object-boundary in polar coordinate is proposed to solve the positioningproblem where the target rotating around the z axis. The algorithm has the advantageof high speed and precision.(3) Target tracking and grasping. A method based on servomotor+synchronous-conveyor is proposed for multiple objects’ tracking and grasping. Byutilizing the encoder of the servomotor, duplication or omission of multiple objects’identifying is avoided. Besides, Newton’s dichotomy method is used for object’sgrasping position calculation. The method fulfills the high frequency pick-and-placeoperation in real time.In the respect of robot kinematic calibration, a fast calibration method using “monocular vision+laser displacement sensor” as is investigated, which can becarried out in fieldwork and improves the robot precision obviously. The major meritscan be summarized in brief.(1) Robot error modeling. An error modeling technique is developed for thekinematic calibration of Delta robot. The model allows the geometric errors affectingthe position and orientation accuracy of the end-effector to be separated into thecompensable and the uncompensable. Error sensitivity analysis is made to reach theconclusion that boundary error is the most sensitive, thus reveals that themeasurement of the error in direction x and y in whole space and the error in directionz at some particular points is enough for the identification of the geometric errorparameter, which provides theoretical basis for error measurement and parameteridentification.(2) Error measurement and parameter identification. By utilizing P4P method,the error of the robot end-effctor in direction x and y is measured in the mode that theCCD camera is moving with the end-effector and the calibration target is fixed underthe workspace. Besides, by utilizing laser displacement sensor, error in direction z ismeasured. Then the parameters can be identified effectively. The positioningaccuracy is upgraded to0.1mm from1mm after the compensation.A monocular vision measurement based method is proposed in the respect of theintegrated calibration of vision system-robot-production line. The rotation about zaxis between vision system and the production line system is calibrated by utilizingthe conveyor’s distance property. Then the calibration of robot with production line isrealized by using P4P method with the help of a simple calibration target. Theproposed method brings adequate guarantee for the overall accuracy of the robotsystem.The outcome has important theoretical meaning and practical value in the visionbased robot technology and engineering application of parallel robot.更多还原