基于视觉技术的机器人抓取目标识别与定位  

Recognition and location of robot grasping target based on vision technology

作　　者：王丽荣[1] Wang Lirong(School of Electronic Information,Beijing Information Technology College,Beijing,100015,China)

机构地区：[1]北京信息职业技术学院电子信息学院,北京100015

出　　处：《机械设计与制造工程》2021年第10期33-36,共4页Machine Design and Manufacturing Engineering

摘　　要：针对传统机器人目标抓取过程中识别精度低的问题,提出了一种改进SSD网络的目标识别与定位方法。首先,针对机器人在小目标检测上精度低的问题,将特征金字塔与CNN神经网络结合,实现高层特征与底层特征提取的联合;其次,针对工业机器人数据样本小、不利于识别分析的问题,采用移动、旋转、增加噪声干扰、消除干扰数据等方式对数据集进行增强;最后,对改进的SSD网络进行验证。结果表明,相较于SS+SIFT+SVM算法、SS+HOG+SVM算法,改进的SSD网络可有效识别和定位需要抓取的目标。Aiming at the problem of low recognition accuracy in the process of traditional robot target grasping,a target recognition and location method based on improved SSD network is proposed.Firstly,aiming at the low accuracy of robot in small target detection,the feature pyramid is combined with CNN neural network to realize the strong combination of high-level feature and low-level feature extraction.Secondly,aiming at the problem that the data sample of industrial robot is small and not conducive to identification and analysis,the data set is enhanced by moving,rotating,increasing noise interference and eliminating interference data;Finally,the improved SSD network is verified.The results show that compared with SS+SIFT+SVM algorithm and SS+HOG+SVM algorithm,the improved SSD network can effectively identify and locate the targets to be captured.

关 键 词：视觉技术 机器人 目标识别 SSD网络 

分 类 号：TP399[自动化与计算机技术—计算机应用技术] TH115[自动化与计算机技术—计算机科学与技术]