基于单目视觉前馈的机械臂目标快速三维瞄准方法  

Fast 3D Aiming Method of Manipulators Based on Monocular Visual Feedforward

作　　者：陈晓鹏 徐鹏 王占涛[2] 邢程 邱钰涵 CHEN Xiaopeng;XU Peng;WANG Zhantao;XING Cheng;QIU Yuhan(School of Mechatronical Engineering,Beijing Institute of Technology,Beijing 100081,China;Beijing Aerospace Institute for Metrology and Measurement Technology,Beijing 100076,China)

机构地区：[1]北京理工大学机电学院,北京100081 [2]北京航天计量测试技术研究所,北京100076

出　　处：《郑州大学学报(工学版)》2025年第2期11-18,共8页Journal of Zhengzhou University(Engineering Science)

基　　金：国家自然科学基金资助项目(61973031);北京理工大学科技创新计划(2024CX06013)。

摘　　要：针对基于单目视觉的机械臂目标瞄准缺少三维深度信息和收敛速度慢等问题,提出了一种单目视觉前馈的六自由度机械臂快速瞄准方法。首先,对YOLOv4的CSPDarknet网络进行裁剪,减少Conv2d_BN_Mish单元数量,简化特征提取网络复杂度,得到Lite YOLOv4以此提高目标检测速度;其次,根据从单目图像中检测得到的目标像素坐标,求取目标像素坐标的反投影归一化坐标,得到激光瞄准器期望的相对三维瞄准射线方程;再次,提出多点标定法实现单目相机、激光瞄准器以及机械臂坐标系的外参数标定,并根据外参数标定结果与激光瞄准器的期望三维射线方程计算机械臂各个关节的期望姿态;最后,基于机械臂的直接位置控制代替图像视觉伺服,加快收敛速度,实现基于单目视觉前馈的机械臂目标快速三维瞄准。经实验验证,所提方法单目平均瞄准响应时间仅为0.611 s,同时目标瞄准成功率达95.238%,与传统基于图像的视觉伺服的方法相比提高了4.762百分点。To address the issues of the lack of three-dimensional depth information and slow convergence speed in monocular vision-based manipulator aiming,a fast aiming algorithm based on monocular visual feedforward approach for a 6D of robotic arm was proposed.First of all,the CSPDarknet of YOLOv4 was cropped through reducing the number of the Cov2D_BN_Mish units and simplifying the complexity of the backbone network so as to form a Lite YOLOv4 algorithm to accelerate object detection speed.Then,the pixel coordinates of the target object detected from monocular images were inversely projected to a 3D ray emitted from the camera center.The 3D ray was exactly the target ray for the aiming laser.What was more,a multiple common points based calibration approach was proposed to obtain the extrinsic parameters of the monocular camera and the aiming laser,and the expected joint poses of the robotic arm were then deducted based on the expected 3D ray coordinates.Finally,direct position control was used to replace visual servoing control based on the expected joint poses to accelerate convergence speed,and realize the monocular visual feedforward based fast 3D aiming for robotic arms.Experiments verified that the average monocular aiming time was 0.611 second.The aiming success rate was 95.238%,which was 4.762 percentage points higher than traditional image based visual servoing.

关 键 词：单目视觉 视觉前馈 自主瞄准 深度学习 机械臂控制 

分 类 号：TP242.6[自动化与计算机技术—检测技术与自动化装置] TP249[自动化与计算机技术—控制科学与工程]