手势识别引导机械臂的零件智能装配方法  

Intelligent Assembly Method of Parts with Robotic Arm Based on Hand Gesture Recognition

作　　者：李迅波 王瑜[1] 凡雷雷 茹元博 孙佳宁 高翔 范俊飞[3] LI Xunbo;WANG Yu;FAN Leilei;RU Yuanbo;SUN Jianing;GAO Xiang;FAN Junfei(School of Mechanical and Electrical Engineering,University of Electronic Science and Technology of China,Chengdu 611731,China;UESTC-Guangxi Institute of Intelligent Manufacturing Technology,Liuzhou 545003,China;Liuzhou Jinjing Group Co.,Ltd.,Liuzhou 545006,China)

机构地区：[1]电子科技大学机械与电气工程学院,四川成都611731 [2]电子科技大学广西智能制造产业技术研究院,广西柳州545003 [3]广西津晶集团有限责任公司,广西柳州545006

出　　处：《电子机械工程》2022年第5期6-10,22,共6页Electro-Mechanical Engineering

基　　金：广西工业互联网和两化融合项目(桂工2021-37号)。

摘　　要：文中利用手势识别方法,以示教编码的模式完成零件自动抓取的动作,防止装配过程中出现错装、漏装等问题,满足零件柔性装配多样化的需求。首先提出利用SSD模型实现零件抓取位置目标检测方法,根据SSD网络给出当前输入图像的零件种类与位置,缩小零件抓取位置分析范围;然后通过整合空间金字塔池化方法与传统特征提取卷积网络,利用传统的边框检测法,设计了物体角度的抓取参考矩形框方法来代替滑动窗口生成方法,以提高检测速度;最后结合YOLOv3网络进行手势识别,引导机械臂进行零件的智能装配。Automatic grasping of parts is achieved by demonstration coding mode based on hand gesture recognition in this paper to avoid misassembly and missing assembly to meet flexible assembly diversity requirements of parts.A target detection method of detecting the grasping position of parts based on SSD model is proposed.Firstly,the part type and position of the current input image are given through the SSD network and the analysis range of the part grasping position is reduced.Then,by integrating the spatial pyramid pooling method with the traditional feature extraction convolution network,adopting the traditional edge detection method,the method of reference rectangular frame for grasping considering object angle is designed to replace the sliding window generation method,which greatly improves the detection speed.Finally,the hand gesture recognition is performed based on YOLOv3 network and guide the robotic arm to carry out intelligent assembly of parts.

关 键 词：智能装配 SSD模型 共享卷积 手势识别 

分 类 号：TH161.7[机械工程—机械制造及自动化]