基于图像梯度向量映射的机械臂姿态估计方法  

Robot arm pose prediction method based on image gradient vector mapping

作　　者：丁宇航 陈震[1] DING Yuhang;CHEN Zhen(State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学海洋工程国家重点实验室,上海200240

出　　处：《中国舰船研究》2023年第5期251-259,共9页Chinese Journal of Ship Research

摘　　要：[目的]为了提高机械臂姿态估计精度和实时性,提出基于RGB图像梯度特征向量映射的机械臂姿态重建方法。[方法]首先,采用方向梯度直方图算法(HOG)计算系列机械臂图像纹理梯度特征,再通过训练深度神经网络(DNN)建立图像特征向量与机械臂关节角度向量之间的映射关系;然后,使用用于预训练的向量映射模型对机械臂运动帧图像进行快速姿态估计;最后,采用合成数据技术生成模型的训练和测试数据集。[结果]试验结果显示,目标机械臂3个关节的角度预测误差平均值为2.92°,单帧图像姿态估计耗时0.08 s。[结论]研究表明,所提方法具有较好的预测速度和精度,仅利用RGB图像信息可实现端到端的机械臂姿态估计。[Objectives]In order to solve the problems of the complexity of the existing robot arm pose prediction algorithm model and its over-reliance on the parameters of the camera and robot,a new robot arm pose prediction method based on RGB image gradient vector mapping is proposed.[Methods]First,a series of robot arm image texture gradient features is calculated based on the Histogram of Oriented Gradient(HOG)algorithm.The mapping relationship between the image features and joint angles of the robot arm is then established by training Deep Neural Networks(DNNs).Finally,the pre-trained vector mapping model is used to quickly predict the pose of the robot arm in a motion frame image.The training and test datasets of the model are generated by synthetic data techniques.[Results]The results show that the average error of the angle prediction of the three joints of the target robot arm is 2.92°,and the pose prediction time of a single image is about 0.08 s.[Conclusions]The results show that the proposed pose prediction method has better prediction speed and accuracy,and only uses RGB image information to achieve end-to-end pose prediction.

关 键 词：焊接机器人 机械臂姿态估计 机器视觉 向量映射模型 方向梯度 深度神经网络 

分 类 号：U671.99[交通运输工程—船舶及航道工程]