为了满足视觉机器人能够精准抓取平面零件的需求,提出一种加入深度学习算法的零件识别与定位方法.首先,利用YOLOv4-tiny目标检测算法对目标物体进行识别,并提取出感兴趣区域(Region of Interest,ROI)送入PSPnet语义分割网络中进一步提取ROI.然后,将ROI区域进行亚像素级的模板匹配,并计算目标物体的深度信息.在目标物体中心坐标求解中,以ROI区域的最大内接圆的圆心作为目标物体的中心.最后,利用D-H法对机器人进行运动学解算,并进行抓取试验.实验结果表明:该方法的深度误差率大约为 0.72%,视觉机器人抓取零件成功率达到91%,具有较高的定位精度和抓取成功率,可以满足实际工业分拣搬运需求.
Research on Part Recognition and Grasping Method of Vision Robot
In order to meet the need of grasping planar parts accurately by visual robot,an algorithm based on deep learning in parts identification and location is proposed.Firstly,YOLOv4-tiny objects detection algorithm//was used to//identify the target and extract the region of interest(ROI),and the ROI was sent to the PSPnet semantic segmentation network to extract it again.Then,the ROI region was matched with sub-pixel templates,and the depth information of the target object was calculated.To solve the center coordinate of the target object,a maximum inscribed circle in the ROI region was used as the center of the target object.Finally,we took full advantage of the D-H method to the inverse kinematics of robots,and performed the grab tests.Ex-perimental results show that the depth error rates of this method are about 0.72%,and the success rates of visual robot in grasping parts reach 91%.This method has high positioning accuracy and grasping success rate,which can meet the actual needs of sort-ing and handling in manufacturing.
万方数据
维普
