面向工业应用的机器人手眼标定与物体定位

发布时间：2018-06-03 00:21

本文选题：工业机器人 + 视觉系统　；参考：《浙江大学》2016年硕士论文

【摘要】：为了满足制造业转型升级发展的需求,集成了视觉系统的智能工业机器人在现代化工厂中被越来越多地使用。为了适应柔性制造中快速部署生产的发展特点,并满足精确识别定位物体的作业需求,本文对工业机器人的手眼标定和物体识别定位问题进行了探索和研究。本文的研究内容和成果主要包括以下几个方面：1.设计并实现了一种在线自动化的手眼标定系统。该系统在执行标定算法的同时,可以自动的采集标定数据,基于摄像机成像模型的标定板运动空间规划保证了在采集数据时标定板出现在摄像机视野范围内,使系统获取有效的标定数据。手眼标定算法采用线性化算法,保证了在线计算的实时性。同时,设计了有效的系统流程控制标定的开始和结束,保证采集到充足的标定数据,以消除观测误差的影响。实验证明,本文设计的自动化标定方法可以得到收敛的标定结果,且整个标定过程仅耗时15min。2.提出并实现了基于最小化重投影误差的手眼标定优化算法。该算法将摄像机成像模型和机器人手眼模型作为一个整体进行建模,采用图像中的棋盘格角点的像坐标作为直接观测数据,在像素空间对模型参数进行优化,将手眼变换矩阵的估计误差转换为棋盘格角点的重投影误差,以最小化重投影误差作为优化目标。同时为了求解含有两部分未知数的优化问题,采用了迭代优化方法。实验证明,该算法可以实现0.873mm的相对标定精度。3.面向工业应用设计并实现了一种采用ORB(Oriented FAST and Rotated BRIEF)特征的物体识别算法和基于物体局部形状特征的物体定位算法。物体识别采用基于特征点匹配的方法,选取旋转不变性和实时性较好的ORB特征。得到特征匹配关系后,使用RANSAC计算感知图像和模板图像之间的单应矩阵,完成物体的初步定位。在此基础上,提出了基于物体局部形状特征的定位优化算法,对物体进行重定位,校正初定位结果。实验证明,该识别算法可以实现工业环境下电路板类物体的快速稳定识别,定位优化算法可以将相对定位误差由0.5658mm降到0.1770mm。
[Abstract]:In order to meet the needs of the transformation and upgrading of manufacturing industry, intelligent industrial robots integrated with visual systems have been used more and more in modern chemical plants. In order to adapt to the development characteristics of rapid deployment of production in flexible manufacturing and to meet the operational requirements of accurate identification of positioning objects, this paper explores and studies the hand-eye calibration and object recognition and positioning of industrial robots. The research contents and achievements of this paper mainly include the following several aspects: 1. An on-line automatic hand-eye calibration system is designed and implemented. The system can automatically collect calibration data while performing calibration algorithm. The moving space planning of calibration board based on camera imaging model ensures that the calibration board appears in the camera field of vision when the data is collected. The system can obtain effective calibration data. Hand-eye calibration algorithm uses linearization algorithm to ensure the real-time of online computing. At the same time, an effective system flow control is designed to control the start and end of calibration to ensure that sufficient calibration data are collected to eliminate the influence of observation errors. Experimental results show that the proposed automatic calibration method can obtain convergent calibration results, and the whole calibration process takes only 15 min. 2. An optimal hand-eye calibration algorithm based on minimizing reprojection error is proposed and implemented. In this algorithm, the camera imaging model and the robot hand-eye model are modeled as a whole, and the image coordinates of the chessboard corner point in the image are used as the direct observation data to optimize the model parameters in the pixel space. The estimation error of the hand-eye transformation matrix is transformed into the reprojection error of the chessboard grid corner, and the optimization objective is to minimize the reprojection error. At the same time, an iterative optimization method is used to solve the optimization problem with two parts unknown numbers. Experiments show that the algorithm can achieve the relative calibration accuracy of 0.873mm. An object recognition algorithm based on ORB(Oriented FAST and Rotated BRIEF) feature and an object location algorithm based on local shape feature are designed and implemented for industrial applications. Object recognition is based on feature point matching, and ORB features with rotation invariance and good real-time performance are selected. After the feature matching relationship is obtained, the monoclinic matrix between the perceptual image and the template image is calculated by using RANSAC to complete the initial location of the object. On this basis, an optimization algorithm based on the local shape features of the object is proposed to relocate the object and correct the initial location results. Experiments show that the algorithm can realize fast and stable recognition of PCB objects in industrial environment, and the location optimization algorithm can reduce the relative positioning error from 0.5658mm to 0.1770 mm.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41;TP242

【参考文献】