餐具视觉分拣系统的设计与实现-《集美大学学报（自然科学版）》

文章信息/Info

作者:: 曾伟民1; 郑东强1; 周海峰2; 李波1; 王云超1; 1.集美大学海洋装备与机械工程学院，福建厦门 361021；2.集美大学轮机工程学院，福建厦门361021

Author(s):: ZENG Weimin1; ZHENG Dongqiang1; ZHOU Haifeng2; LI Bo1; WANG Yunchao1; 1.School of Marine Equipment and Mechanical Engineering，Jimei University，Xiamen 361021,China;2.School of Marine Engineering,Jimei University，Xiamen 361021,China

Keywords:: cleaning of tableware; visual sorting; hand-eye calibration; deep learning

摘要:: 基于PyTorch深度学习框架和OpenCV开源计算机视觉库，利用CenterNet目标检测网络对餐具进行检测,实现无人化分拣装箱。工控机与工业相机通过以太网连接，同时与可编辑逻辑控制器(PLC）和机器人通过Modbus TCP通信。通过相机像素坐标与机器人的世界坐标进行标定，实现了餐具的机器人坐标定位，并控制机器人在分拣线上完成餐具的动态分拣装箱。试验结果表明：当传送速度为5 m·min-1且分拣定位精度≤2 mm时，单台机器人每小时可分拣餐具1800个以上，正确率为100%，达到实用要求。

Abstract:: This article uses the PyTorch deep learning framework and OpenCV open-source computer vision library to detect tableware objects using the CenterNet object detection network and to realize unmanned sorting and packing.The industrial computer was connected to the industrial camera via Ethernet,and Modbus TCP communication was used between the programmable logic controller （PLC） and the robot.By calibrating the camera pixel coordinates with the robot’s world coordinates,robot coordinate positioning of tableware was achieved,and the robot was controlled to dynamically sort and pack tableware on the sorting line.The experimental results show that with the conveying speed of 5 m·min-1and the sorting accuracy under 2 mm, more than 1800 tableware pieces can be sorted per hour with a 100% accuracy rate, which meets the practical requirements.