| 摘要: |
| 针对工业机器人的抓取技术普遍采用离线示教的方式，目标物的改变就会导致抓取不准的问题，研究了一种面向机器人抓取的双目视觉单目多目标检测方法，强化了机器人的抓取能力；首先搭建机器人抓取视觉检测系统，系统采用双目测距模型定位深度，单步多目标检测器定位像素坐标；通过相机标定和eye-to-Hand手眼标定将图像中的像素坐标以及双目测距模型的深度坐标转换为机器人基坐标的位姿，建立机器人运动学模型完成机器人基坐标到机器人末端坐标的转换，通过ROS实现上位机与机器人通讯，并将定位结果发送给ROS；经过多次实验，表明方法的目标检测误差在35 mm以内，深度误差在12 mm之内，具有很好的实用推广价值。 |
| 关键词: 单步多目标检测 双目视觉 机器人抓取；相机标定 |
| DOI： |
| 分类号: |
| 基金项目: |
|
| Application of Binocular Vision Single Step Multi-target Detection Method for Robot Grasping |
|
LI Shi-pei，LI Chun-lin, HAN Jia-bu, ZHU Xin-long
|
|
School of Mechanical and Automobile Engineering,Shanghai University of Engineering Science，Shanghai 201600,China
|
| Abstract: |
| Grasping technology of industrial robots is generally adopted in the way of off-line teaching, the change of the object will lead to the problem of inaccurate grasping. This paper studies a binocular vision monocular multi-object detection method for robot grasping, which enhances the robot′s grasping ability. Firstly, the robot grasping vision detection system is built. This system adopts binocular ranging model to locate depth and single-step multi-target detector to locate pixel coordinates. Through the camera calibration and eye-to-hand hand-eye calibration image pixel coordinates and the depth of the binocular distance measurement model of the coordinate transformation for the robot base coordinate position, robot kinematics model is set up to complete thetransformation from robot base coordinate to robot end coordinate, by ROS, superordination machine and the robot localization is realized, and the result is sent to ROS. After many experiments, it is shown that the target detection error of this method is within 3.5 mm and the depth error is within 1.2 mm, which has a good practical popularization value. |
| Key words: single-step multi-target detection binocular vision robot grasping camera calibration |
