| 李乐乐， 张团善， 马浩然， 张越.基于Yolov4 Tiny与RANSAC算法的管纱识别抓取系统[J].轻工机械,2021,39(4):68-73 |
| 基于Yolov4 Tiny与RANSAC算法的管纱识别抓取系统 |
| Yarn Recognition and Grasping System Based on Yolov4 Tiny and Ransac Algorithm |
| |
| DOI：10.3969/j.issn.1005 2895.2021.04.013 |
| 中文关键词: 机器视觉 RGB D图像 RANSAC算法 Yolov4 Tiny模型 点云配准 |
| 英文关键词:machine vision RGB D images Ransac algorithm Yolov4 Tiny model point cloud registration |
| 基金项目:西安市现代智能纺织装备重点实验室项目（2019220614SYS021CG043）。 |
|
| 摘要点击次数: 546 |
| 全文下载次数: 635 |
| 中文摘要: |
| 现了管纱的检测、定位及抓取任务。首先利用深度相机获取RGB D图像，通过训练深度学习网络Yolov4 Tiny，生成预测结果；然后利用预测框信息将原始点云进行裁剪，采用快速点特征直方图与RANSAC算法进行配准；最后利用手眼标定的方法将深度图像坐标信息转换到机械臂坐标系下完成抓取。实验结果表明：系统管纱平均抓取成功率达到65%，在非结构化的环境中具有良好的抓取结果，满足管纱抓取的实际生产需求。 |
| 英文摘要: |
| To solve the non automatic problem that the winding machine relies on manual yarn loading, a recognition and grasping system based on Yolov4 Tiny object detection model was proposed to realize the detection, positioning and grasping tasks of the cop. Firstly, the depth cameras were used to obtain RGB D images, and the forecast results were generated through training Yolov4 Tiny deep learning network. Then the original point cloud was clipped with the prediction information, and the fast point feature histogram and RANSAC algorithm were used for registration. Finally the method of hand eye calibration was used to convert the depth image coordinates information to mechanical arm coordinate system to complete the capture. The experimental results show that the average successful grasping rate of the proposed system reaches 65%, which has good grasping results in the unstructured environment and meets the actual production requirements of grasping cop. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
