摘要
针对机械手抓取物体大多以指定位置抓取特定物体的方式及柔性差的问题,提出利用基于深度学习方式的目标检测算法对物体进行识别。通过双目视觉算法检测物体所在的空间位置,利用D-H法进行机械手的坐标解算,从而实现物体的抓取。根据实际需求,采用实时性较好的YOLOv4目标检测算法与OpenCV中的立体匹配算法SGBM相结合的方式实现目标定位检测,并且通过租用云端服务器来训练神经网络和运行程序的方式降低本地硬件要求。实验结果表明:该机械手抓取物体的成功率达到了84%,验证了该方法具有较好的准确性,基本满足智能制造中的实际需求。
Aiming to solve the problems that most of manipulator grasping objects in the specified position and poor flexibility, a target detection algorithm based on deep learning was proposed to identify objects. Binocular vision algorithm was used to detect the spatial position of objects, and D-H method was used to solve the coordinate of manipulator, so the object grasping was realized. According to actual demand, target location detection was realized by combining the good real-time target detection algorithm YOLOv4 with the stereo matching algorithm SGBM in OpenCV, and the cloud server was rented to train the neural network and run programs, so the local hardware requirements was reduced. The experimental results show that the success rate of the manipulator grasping objects reaches 84%, which verifies that the method has good accuracy and basically meets the actual needs of intelligent manufacturing.
作者
袁斌
王辉
王伟博
吴瑞明
YUAN Bin;WANG Hui;WANG Weibo;WU Ruiming(School of Mechanical and Energy Engineering,Zhejiang University of Science and Technology,Hangzhou Zhejiang 310023,China)
出处
《机床与液压》
北大核心
2021年第23期43-47,共5页
Machine Tool & Hydraulics
基金
浙江省基础公益研究计划(LGG19E050003)。
