摘要
提出了一种基于RGB-D图像的三维物体检测与抓取方法。该方法主要实现RGB-D图像的模板匹配,对匹配结果进行聚类、评估和非极大值抑制,物体定位,并在三维点云中分割出平滑曲面,从而计算机器人的抓取位置与姿态。实验结果表明:该方法可在杂乱环境下对目标进行定位,并引导机器人抓取,抓取成功率达到89%。
This paper proposes a 3D object detection and grasping pose generation method based on RGB-D images. The approach starts with template matching and clusters the matched templates of similar spatial location. Then each cluster is evaluated based on a scoring function, and non-maximum suppression is adopted to remove duplicate results based on the scores. Then the feasible grasping poses is computed by extracting smooth surface from 3D point clouds. The experimental results show the approach can robustly detect objects in cluttered environment and guides robot to accomplish pick-and-place tasks, achieving an average success rate of 89%.
出处
《机械工程与自动化》
2017年第5期28-30,共3页
Mechanical Engineering & Automation
基金
广东省重大科技专项(2014B090919002
2016B0911006)
