摘要
针对机器人工业抓取场景中,堆叠的工件相互遮挡,难以识别的问题,提出一种基于聚类网格法的自适应线性推动策略AC-Grid。融合二维图像与点云高度渲染信息,根据抓取场景内工件的散乱堆叠情况分析出可靠的几何特征,为机器人优化出一条合理有效的工件推动路线。在V-REP仿真环境中制作80组“PushTD”系列的模拟场景数据集,对比实验结果表明,AC-Grid推动策略在仿真场景下最高能使平均目标匹配识别度提高至39.6%,在实际场景中能达到16.4%,在不同场景中均能起到显著分离和目标识别度提升的作用。
In the robotic industrial grasping scene,the stacked work-pieces obscure each other,making it difficult to identify.To solve the problem,an adaptive linear driving strategy AC-Grid based on the clustering grid method was demonstrated.Two-dimensional images and point cloud height rendering information were combined.Reliable geometric features were analyzed based on the scattered stacking of work-pieces in the captured scene.Therefore,a reasonable and effective work-piece pushing route for the robot was optimized.In the V-REP simulation environment,80 sets of simulation scene datasets of the“PushTD”series were produced.Results of comparative experiments show that the AC-Grid promotion strategy can increase the average target matching recognition degree to 39.6%in the simulation scene,which can reach 16.4%in real situation.The simulation results show the proposed method possess promotion in separation and target recognition in different scenarios.
作者
赵有港
张宏
徐刚
许允款
曾晶
ZHAO You-gang;ZHANG Hong;XU Gang;XU Yun-kuan;ZENG Jing(School of Mechanical Engineering,Taiyuan University of Science and Technology,Taiyuan 030024,China;Computer Vision Lab,Ningbo Institute of Materials Technology and Engineering,Chinese Academy of Sciences,Ningbo 315000,China)
出处
《计算机工程与设计》
北大核心
2023年第4期1242-1249,共8页
Computer Engineering and Design
基金
宁波市科技创新2025重大专项基金项目(2020Z013)。
关键词
机器人抓取
聚类网格
推动策略
图像处理
目标分离
目标识别
模拟场景数据集
robot grabbing
clustering grid
pushing strategy
image processing
target separation
target recognition
simulation scene datasets
