【作者】 丁宁； 路士州； 许超； 王延刚； 毛维顺； 刘鑫；

【Author】 DING Ning;LU Shizhou;XU Chao;WANG Yangang;MAO Weishun;LIU Xin;School of Mechanical,Electrical & Information Engineering,Shandong University;School of Electronic Engineering and Computer Science,Peking University;Weihai Huadong Automation Co.,Ltd.;

【机构】 山东大学机电与信息工程学院； 北京大学信息科学技术学院； 威海华东数控股份有限公司；

【摘要】 针对工业流水线上外形相似、区分度较低的多种阶梯轴类零件,提出了一种利用单目相机对多个目标同时进行分类和定位的算法。工业相机拍摄到的图像经过滤波降噪、边缘提取、形态学处理等步骤后被分割成多个含有零件的感兴趣区域,沿区域中零件的长轴方向采样多条直径归一化后形成特征向量,并与模板向量进行匹配,进而完成分类;对于零件的实际位置坐标,通过内参矩阵的标定与外参矩阵的求解相结合便可求得。实验结果表明该分类与定位算法具有良好的准确性。系统及算法通过开源计算机视觉库OpenCV实现,可以在较低的硬件配置下稳定运行,十分方便工业流水线部署。更多还原

【Abstract】 In this paper,an algorithm to classify and localize multiple targets using monocular camera is proposed for step-shaft parts with similar shape and low discrimination in industrial pipeline.After filtering and denoising,edge extraction,morphology processing and other steps,the image captured by the industrial camera is divided into multiple regions of interest.Several diameters are sampled along the axial direction of the parts in the region to form the feature vector,and then it matches with template vectors to complete the classification.As for the coordinates localization of the parts,it can be obtained by the calibration of intrinsic parameter together with the solution of extrinsic parameter.The experimental result shows that the algorithm has good accuracy.The system and algorithm are implemented by open-sourced library OpenCV,which can run stably under low hardware configuration,and is very convenient for industrial deployment.更多还原