基于物体逻辑状态推理的未知物体视觉分割

发布时间：2018-05-28 02:43

  本文选题：未知物体分割 + 视觉分割　； 参考：《机器人》2017年04期

【摘要】：为提高机器人的视觉感知能力,提出了一种新颖的基于物体逻辑状态推理的未知物体视觉分割方法.在语义层定义物体逻辑状态空间,根据机器人抓取动作反馈对物体逻辑状态进行推理.在数据层利用RGB-D摄像头采集生成场景3维着色点云,基于物体放置于支撑平面的假设,对所有可能的物体点进行空间聚类和分割,得到初始未知物体集并生成物体初始逻辑状态表示.当物体逻辑状态发生变化时,根据设定规则结合点集运算对物体点云进行重新分割,物体点云集的变化又用于指导物体逻辑状态空间的更新.在7自由度移动机械手系统上,进行了真实积木模拟环境下未知物体的视觉分割与抓取实验,实验结果证实本文方法可以有效提升机器人的视觉感知能力.
[Abstract]:In order to improve the visual perception ability of robot, a novel object segmentation method based on object logic state reasoning is proposed. The object logic state space is defined at the semantic level, and the object logic state is inferred according to the robot grab action feedback. In the data layer, RGB-D camera is used to capture and generate 3D colored point clouds. Based on the assumption that objects are placed in the support plane, all possible object points are clustered and segmented in space. The initial set of unknown objects is obtained and the initial logical state representation of the object is generated. When the logical state of the object changes, the point cloud is resegmented according to the set of set rules, and the change of the point cloud is used to guide the updating of the logical state space of the object. In the 7 DOF mobile manipulator system, the experiments of visual segmentation and grasping of unknown objects in real building block simulation environment are carried out. The experimental results show that the proposed method can effectively improve the visual perception ability of the robot.
【作者单位】： 扬州大学水利与能源动力工程学院;东南大学仪器科学与工程学院;
【基金】：江苏省博士后基金(1601006C) 国家自然科学基金(61325018,61272379)
【分类号】：TP242;TP391.41
，

本文编号：1944942