中国科学院机构知识库网格
Chinese Academy of Sciences Institutional Repositories Grid
装配机器人系统的工件定位和抓取装配研究

文献类型:学位论文

作者赵广涛
学位类别工学博士
答辩日期2011-04-20
授予单位中国科学院研究生院
授予地点中国科学院自动化研究所
导师乔红
关键词装配机器人 目标检测 三维工件定位 图像矩 抓取点提取 Assembly robot Target recognition Work-piece location Image moment Extraction of the grasp point
其他题名Research on the location and Grasp of the Work-piece based on the Assembly Robotic System
学位专业控制理论与控制工程
中文摘要装配几乎是所有工业生产的必要环节,也是工业生产花费最大的环节之一,在制造领域占有重要地位。装配机器人的出现使得装配系统在降低成本的同时极大提高了系统的运行效率。而近年来,基于视觉技术的装配机器人系统的出现进一步提高了装配操作的灵活性,并逐渐成为机器人装配领域的一个重要发展趋势。它通过视觉系统自动对工件进行检测、识别和定位,从而能够适用于不同种类工件的混合装配系统,并已在汽车、船舶、航空航天和电子仪表等行业得到初步应用。 目前,阻碍基于视觉的装配系统实用化的一个主要问题是其相对低下的运行效率,存在检测实时性要求高、三维工件姿态变化所带来的定位不准确和有效抓取点难提取等方面的问题。因此研究快速且稳定的基于视觉的工件检测、三维准确定位和有效抓取装配方法,对于提高装配系统的整体性能具有重要的理论和应用意义。针对上面所述的问题,本文开展了下面的研究工作: 1、针对工业装配现场零件检测实时性要求高的特点,提出了一种在仿射变换条件下利用部分Hausdorff距离进行零件检测的高效算法。针对目标的高效搜索问题,首先将不同工件图像的仿射变换空间转换到工件成像空间,然后基于目标图像的一阶、二阶图像矩和装配平台的固有特性将搜索空间集中在两维空间内,从而降低了模板搜索的复杂度,提高了工件的识别速度。通过与其它两种方法的实验比较,验证了该方法的快速性和有效性。 2、针对三维工件定位中摄像机位置和工件姿态变化所带来的定位困难,在深入分析工件图像特征点与工件位姿对应关系的基础上,提出了基于工件图像统计特征的初定位和快速误差修正的精确定位方法。该方法利用工件的代数矩特征和工件自身的误差校正模型,计算得到了满足工业装配要求的高精度定位结果,相关实验验证了该定位方法的有效性。 3、针对已知或未知工件的抓取问题,在深入分析抓取点确定过程中需遵循的力封闭准则和抓取稳定性准则的基础上,建立了与抓取稳定性、装配有效性和先验知识有关的能量函数,从而将抓取点的确定问题转化为能量函数的优化问题。通过该能量函数在工件边缘上进行最优搜索,得到了满足要求的抓取点。对工件上不同抓取点的能量函数的分析以及最优抓取点的提取实验,验证了该算法的有效性。 4、设计并搭建了基于单目视觉的工业装配机器人系统。针对视觉系统的测量精度问题,提出了基于交比不变性的摄像机标定方法,提高了测量精度。针对单目视觉的控制问题,设计并实现了简化的基于二维图像的视觉控制结构,基于图像雅可比矩阵的正定性分析,证明了该视觉控制结构的稳定性。最后,设计了有效的抓取和装配策略,实现了对三维工件的高精度抓取和装配研究。 工业装配现场测试实验验证了本文所提出的基于视觉的工件检测、定位和抓取装配方法的有效性。
英文摘要The assembly manipulation is nearly the necessary step in current industrial production and costs very much. In the field of assembly robotic system, vision-based systems make the robotic system able to dynamically inspect, recognize and locate the target work-piece and these functions reduce the production cost and improve the flexibility of the assembly manipulation. Currently, vision-based assembly system is applied in the industry such as automobile, astronaut and electronic instrument and so on. Regarding to the current vision-based assembly system, the main problem to hinder the application of the assembly robotic robot is that the operating efficiency of the sytem is low dew to the large computation resulted from the recognition, location and the control structure. The main influencing factors consist of the real-time inspection of the target work-piece, the inaccuracy resulted by posture change of the target 3-D work-piece, the reasonable extraction of grasp position and so on. To improve the whole perfermence of the assembly robotic system, it is very meaningful to research into both improving the performance of recognition and location of three dimensional work-pieces and improving the real-time performance of assembly system. In this dissertation, the issues concerned with assembly robotic system are discussed, including recognition and location of three dimensional work-pieces and the visual control of the assembly robotic system. Firstly, a method for recognition of different work-pieces is proposed based on model matching. The Hausdorff distance is applied to measure matching degree between shape of the work-piece to be recognized and standard models. Through extraction of the first and second order image moments, it can be estimated that are three translation parameters and one rotation parameter between two images of the target work-piece. Therefore, the dimension of searching space is reduced mainly in the remaining two dimensions and the matching speed is greatly improved. Compared with other two recognition methods, the effectiveness is identified for the above recognition method. Secondly, the accurate location of the work-pieces is researched based on the image moments. According to industrial assembly characteristic, shape feature is chosen as classification basis on analyzing the location operation of work-pieces. The types of the work-pieces are divided into two categories: work-piece with isotropy and work-piece with non-isotro...
语种中文
其他标识符200618014628017
源URL[http://ir.ia.ac.cn/handle/173211/6321]  
专题毕业生_博士学位论文
推荐引用方式
GB/T 7714
赵广涛. 装配机器人系统的工件定位和抓取装配研究[D]. 中国科学院自动化研究所. 中国科学院研究生院. 2011.

入库方式: OAI收割

来源:自动化研究所

浏览0
下载0
收藏0
其他版本

除非特别说明,本系统中所有内容都受版权保护,并保留所有权利。