【作者】 赵鹏；

【导师】 李勃； 陈启美；

【作者基本信息】 南京大学 ， 电子与通信工程（专业学位）， 2017， 硕士

【摘要】 工业机器人是“工业4.0”的重要载体,推动着我国制造业转型升级。更加精致的传感器和决策系统可以大大提升机器人的智能化水平,机器人视觉系统便是其中之一。工业级应用要求机器人视觉算法拥有更高适应性、更高精度、更高鲁棒性和更高性能。针对以上需求,文章对图像畸变校正算法、几何特征识别与定位相关算法和Scara机器人视觉标定算法等机器人视觉所需的关键算法进行深入研究,改进了多个相关算法模块,主要包括提高畸变校正的稳定性并简化应用流程,实验测试多种不同亚像素边缘检测算法并进行对比,改进直线检测和圆检测在强干扰状况下的表现,优化模板匹配算法的效率以及提出了多种相机安装方式下的视觉标定算法。文章的主要工作和创新点或特点在于:●提出了基于共线点集的图像畸变校正算法,解决了传统算法中内外参数耦合导致求解结果不稳定的问题,不仅简化了计算并且增加了使用便捷性。●提出了应用Hough变换计算初值然后进行距离加权迭代拟合的算法,Hough变换可提高直线拟合的抗干扰能力,距离加权迭代法可提高直线拟合的精度。●提出改进Haustorff距离和梯度差分相结合的二级判据的模板匹配算法,并对算法的搜索流程进行优化,有效提高了模板匹配的效率,使其满足工业生产的需要。●研究了相机固定安装和相机随动安装两种标定算法,创新性地提出了相机固定安装自动标定的方案,不仅解决了手动标定带来精度损失的问题,而且大大简化了标定流程,实验表明,这种方式的标定误差低于0.2像素。针对复杂多样的工业应用需求,通过整合、改进和封装Scara机器人视觉系统的关键算法可实现快速无码化二次开发,缩短研发周期和提高系统质量。多个项目实践证明,以文章中关键算法为核心的机器人视觉系统应用价值高,市场竞争力强。更多还原

【Abstract】 Industrial robots are the important carrier of "industrial 4.0" which promote China’s manufacturing transformation and upgrading.More sophisticated sensors and decision-making system can greatly enhance the level of robot’s intelligence.The robot vision system is one of them.Industrial applications require robotic vision algorithms with greater adaptability,higher accuracy,higher robustness,and higher performance.In order to meet the above requirements,much common used algorithms such as the image distortion correction algorithm,geometric feature recognition and location correlation algorithm and Scara robot vision calibration algorithm are deeply studied.Some optimized solutions are proposed,such as improving the stability of distortion correction result and simplifying application process.The performance of linear detection and circle detection in strong interference conditions was improved.What’s more,the efficiency of template matching algorithm is improved and the visual calibration algorithm under a variety of camera installation is proposed.The main work,innovation or features of this dissertation are summarized as follows:● To avoid the coupling of internal and external parameters which can lead to the instability of the result,an image distortion correction algorithm based on collinear points is proposed.More than that,the new algorithm increases the convenience of the use.● The sub-pixel edge detection algorithm is studied and analyzed.The method of determining the accuracy of Gaussian fitting method,Zernike moment method and interpolation method is proposed by comparing the fitting distance,and the experimental verification is carried out.● Hough transform algorithm is used calculate the initial value for distance-weighted iterative fitting.Hough transform is proposed to improve the anti-jamming ability of straight line fitting,and the distance-iterative method is used to improve the precision of straight line fitting.● A double threshold maximum interclass variance algorithm can help obtain five different gray levels of the image,and enhance the scene adaptability of the module.● A template matching algorithm which use the improved Haustorff distance and gradient difference to make two-level decisions is proposed.The search flow of the algorithm optimized to improve the efficiency of template matching to meet the needs of industrial production.● After a lot of research,the camera fixed installation automatic calibration algorithm is put forward,which not only solved the problem of loss of accuracy resulted by manual calibration,but also greatly simplifies the calibration process.Many experiments show that the calibration error is less than 0.2 pixels.In order to meet the needs of complex and diverse industrial applications,the key algorithms of Scara robot vision system are integrated,improved and encapsulated.This program can realize rapid secondary development,shorten the development cycle and improve the quality of the system.A number of projects have proved that the robot vision system equipped by algorithms presented in the paper has much application value and strong market competitiveness.更多还原