基于改进YOLOv3的机器人定位抓取研究
Research on Robot Positioning and Grasping Based on Improved YOLOv3
王新庆 1王新 1杨振宇 1李世庆 1邹宇鹏1
作者信息
- 1. 中国石油大学(华东)机电工程学院,青岛 266580
- 折叠
摘要
针对视觉引导定位的机械臂抓取装配任务中对工业零件的定位差、抓取效率低等问题,提出一种基于改进YOLOv3 的智能抓取系统方案,实现工业零件从目标检测到自动化抓取的智能化.首先,为提高对小目标和拥挤目标的检测性能,提出了改进YOLOv3 目标检测网络;其次,对工业零件进行数据采集与训练,实现对零件的目标识别与定位;最后,通过相机标定和手眼标定实现由图像坐标系到世界坐标系的转变,获得被抓物体的世界坐标,对机械臂进行抓取规划,完成目标物体的抓取.实验采用Kinect V2 相机与UR3 六轴协作机械臂组成抓取实验平台,对目标零件进行定位和抓取实验.实验结果表明,改进的YOLOv3 算法提高了对小目标和拥挤目标的检测性能,对零件进行精准的目标定位并抓取.
Abstract
Aiming at the problems of poor positioning and low grasping efficiency of industrial parts in the manipulator grasping assembly task of visual guidance positioning,an improved YOLOv3 intelligent grasp-ing system solution is proposed to realize the intelligent of industrial parts from object detection to automat-ic grasping.First,in order to improve the detection performance of small targets and crowded targets,an im-proved YOLOv3 target detection network is proposed.Secondly,data collection and training are carried out on industrial parts to realize the target recognition and positioning of the parts.Finally,through camera cali-bration and hand-eye calibration,the transformation from the image coordinate system to the world coordi-nate system is realized,the world coordinates of the grasped object are obtained,the grasping plan of the manipulator is carried out,and the grasping of the target object is completed.In the experiment,the Kinect V2 camera and the UR3 six-axis collaborative manipulator were used to form a grasping experiment plat-form,and the positioning and grasping experiments of the target parts were carried out respectively.The ex-perimental results show that the improved YOLOv3 adds the fourth layer of feature-scale target detection,which improves the detection performance of small targets and crowded targets.The grasping system accu-rately locates the parts and successfully grasps them.
关键词
装配/机械臂/目标检测/深度学习/抓取Key words
assembly/manipulator/object detection/deep learning/grasping引用本文复制引用
基金项目
山东省自然科学基金项目(ZR2022MF291)
山东省重大科技创新项目(2017CXGC0902)
出版年
2024
NETL
NSTL
万方数据