【作者】 赵一鸣；

【导师】 胡燕海；

【作者基本信息】 宁波大学 ， 工程硕士（专业学位）， 2022， 硕士

【摘要】 随着时代的发展与科技的进步,制造业已经发生了翻天覆地的变化,智能制造的兴起与发展逐渐成为了不可阻挡的趋势。由柔性自动化设备替代人工作业的现象在制造业已越来越普遍,例如机械臂分拣等。传统机械臂抓取时需要预先将待分拣目标放置于固定位置,因为机械臂的运动轨迹往往是固定的,而当机械臂联合了机器视觉技术后,机械臂抓取时的灵活性将获得极大提升。在工业生产中,机械臂分拣是一个重要的环节。为了解决分拣目标类型具有多样性且分拣情况复杂,人工分拣成本高及人工分拣错误率高等问题,本文联合机械臂和机器视觉技术以提高分拣目标检测轮廓精度以及轮廓识别准确度为目的,对分拣方法进行研究,用以替代人工分拣高效完成分拣任务。首先对分拣平台进行了搭建,机械臂采用了嵌入式机械臂,相机采用海康威视公司的MV-CA060-11GM相机。对机械臂进行了正向运动学分析,确定了其工具坐标系的建立,并对整个分拣平台进行了标定,求得了相机图像的像素坐标与机械臂工具坐标系的转换关系,随后对机械臂进行了逆向运动学分析,确定了分拣时各舵机的旋转角度,完成了分拣前的准备工作。其次,分析了各常用轮廓检测算法的优劣,采用Canny算法进行轮廓提取,为解决Canny算法在光照不均匀、图像噪声增多时,误检率升高的问题,对Canny算法作出了相应改进,同时针对Canny算法高低阈值设置的问题结合遗传算法求得阈值的最优解,实现了图像的轮廓准确提取并完成目标定位。然后根据提取到的轮廓图像,采用Blob分析筛选出图像差异特征并建立轮廓模板库,同时采用Hu不变矩算法对目标轮廓进行匹配,为减小图像尺度差异对算法的影响对模板图像建立多尺度模板,优化了算法搜索匹配的方式以提高匹配执行效率,完成了多类型目标的实时识别。最后整合前期工作,对分拣平台进行总体流程分析,采用Vision Master算法平台与Keil u Vision软件,完成上位机与下位机的设计工作,实现视觉算法配合机械臂控制并最终完成目标分拣。更多还原

【Abstract】 With the development of the times and the progress of science and technology,the manufacturing industry has undergone changes remarkably.The rise and development of intelligent manufacturing has gradually become an irresistible trend.The phenomenon of replacing manual operation by flexible automation equipment is becoming more and more common in the manufacturing industry,such as mechanical arm sorting and so on.For the mechanical arm sorting,the traditional manipulator needs to place the target to be sorted in a fixed position in advance,because the motion trajectory of the manipulator is often fixed.When the manipulator is combined with machine vision technology,the flexibility of the manipulator will be greatly improved.In industrial production,mechanical arm sorting is an important link.To overcome problems like the diversity of sorting target types,complex sorting conditions,high manual sorting cost and high manual sorting error rate,the solution is put forward that mechanical arm is combined with machine vision technology in this thesis.The sorting method is studied for the purpose of improving the contour accuracy of sorting target detection and contour recognition accuracy,so as to replace manual sorting and finish the sorting task efficiently.Firstly,the sorting platform was built,the embedded mechanical arm was used,and the camera was the MV-CA060-11 GM camera of Hikvision.The forward kinematics analysis was carried out on the mechanical arm,make sure the tool coordinate system was established,and calibration of the entire sorting platform,obtained the camera image pixel coordinates and mechanical arm tool coordinate system transformation relations,then the inverse kinematics analysis was carried out on the mechanical arm,determining the sorting the rotation angle of steering gear,the preparation work before sorting was completed.Secondly,the advantages and disadvantages of the commonly used contour detection algorithms were analyzed,and the Canny algorithm was used for contour extraction.In order to solve the problem that the false detection rate of the Canny algorithm increased when the illumination was not uniform and the image noise increased,the Canny algorithm was improved accordingly.At the same time,the problem of setting the high and low threshold of the Canny algorithm was solved by combining the genetic algorithm to obtain the optimal solution of the threshold.Then,according to extracted contour images,Blob analysis was used to screen out image difference features and establish contour template library.Meanwhile,Hu invariant moment algorithm was used to match target contour.The method of searching matching is optimized to improve the efficiency of matching execution,and the real-time recognition of multiple types of targets is completed.Finally,the preliminary work was integrated to analyze the overall process of the sorting platform.The vision master algorithm platform and keil u Vision software were used to complete the design of the upper computer and the lower computer,so as to realize the control of the visual algorithm with the mechanical arm and finally complete the target sorting.更多还原