【作者】 李兵；

【导师】 傅卫平；

【作者基本信息】 西安理工大学 ， 机械电子工程， 2017， 硕士

【摘要】 随着机器人智能化发展,越来越多工业机器人搭载视觉系统。机器人与视觉融合不深仍是目前面临的主要问题。本文针对6自由度串联智能机器人自主轴孔装配任务,基于云台双目相机(Eye to Hand,ETH)和手眼相机(Eye in Hand,EIH)构成的三目视觉系统,分别提出了三目机器视觉系统标定方法、机器人运动学参数标定方法和任意空间孔位姿重建方法,实验结果验证了各方法的有效性。具体内容有:(1)三目机器视觉系统标定。三相机采集同一平面靶标图像,基于ZHANG标定法分别标定其内外参数,通过三视张量约束优化三相机之间位姿关系,以ETH左相机为基础坐标系建立三目机器视觉系统统一坐标系。最后,根据三目视觉定位对标准棋盘纵横方格尺寸进行测量,相对误差在3.2%以内。(2)基于EIH的机器人运动学标定。EIH随关节运动,采集各关节单独运动时固定的平面靶标图像,根据ZHANG标定法计算EIH拍摄各图像处其光心的位置,由各关节单独运动时相机光心轨迹拟合圆得出各关节轴线,分别取各关节最大范围转角处拍摄的两张靶标图像和一张中间处的靶标图像,采用三视张量约束优化各关节轴线方向,即得到各关节旋量,从而可建立机器人指数积运动学模型。最后,通过在机器人不同位姿处末端的EIH拍摄同一棋盘格的两幅图像,利用得到的机器人运动学模型计算前后两处EIH的位姿关系,根据双目重建原理测量出棋盘纵横方格尺寸,相对误差小于3%。(3)基于准线-母线的空间孔位姿重建。取孔横断面上边缘点拟合出孔准线,过孔柱面其他任意点以假定轴线方向作孔母线,采用Levenberg-Marquard法优化多条母线与横断面交点到孔准线代数距离最小得到孔轴线方向,孔准线中心作为孔位置,仿真实验显示斜孔位姿重建相对误差为0.2%,对比实验也得到较高重建精度。最后,提出了一种基于视觉的孔内表面点的检测方法。(4)基于机器人-视觉系统目标孔位置重建综合实验。通过标定三目机器视觉系统,建立装配机器人指数积模型,采用ETH和EIH对工件孔分别进行位姿重建,重建结果进一步验证本文所提出的方法正确性。更多还原

【Abstract】 With the development of robots intelligentialize,more and more industrial robots are equipped with vision system.However,robot and vision deep fusing is still a major issue currently facing.In this paper,based on trinocular vision system consisted of eye to hand(ETH)and eye in hand(EIH),trinocular vision calibration,robot kinematic parameter calibration and arbitrary space hole pose reconstruction methods are proposed respectively for peg and hole auto assembly task by 6-DOF series intelligent robot.The experimental results verify the effectiveness of each method.The specific contents are as follows.(1)Trinocular machine vision system calibration.With a planar target images acquired,three camera internal and external parameters calibrated according to ZHANG calibration method.After that,the relation pose among tri-camera is optimized by trifocal tensor constraint.The unified coordinate system of tri-vision is established with benchmark of the left ETH.Lastly,chessboard horizontal and verical size are measured according to tri-vision loaction theory.The exprement shows the size relative error below 3.2%.(2)Robot kinematic calibration based on EIH.With robot every joint individual movement,EIH collects images of fixed plane target.According to ZHANG calibration method,the camera optic center position at every image shooting is obtained.By trajectory of the optical center fitting circle on each mono-joint rolling,the correspoding joint axis is gained.Whereafter,the every axis is optimized by trifocal tensor constraint consist of two images at its joint minimax rotation angle and the third image at middle,which joint screw is obtained by,then to establish exponential product model of robot.Finally,through two hessboard image achieved by EIH at different pose whose relative pose is calculated,hessboard horizontal and verical size are measured by principle of binocular reconstruction.The experiment shows the size relative error below 3%.(3)Space hole pose reconstruction based on directrix-generatrix.Firstly,hole directrix curve is fitted by marginal points in cross section of hole.With the same direction given of hole axis a generatrix drew through other point in cylinder of hole,the orientation of hole is gained by optimizing the minimun algebraic distance between the directrix and the crossover point between generatrix and cross section.The center of directrix curve can be used as hole location.The simulation experiment of pose reconstruction of inclined hole shows the results of relative error of 0.2%,and physical experiments also get higher reconstruction accuracy.Finally,a method of detecting the surface of the hole is presented based on vision.(4)Comprehensive experiment of target hole pose reconstruction based on the robot-vision system.By calibrating trinocular machine system and establishing the exponential product model of assembly robot,the ETH and EIH are used to reconstruct the workpiece hole pose respectively.Furthermore,the correctness of those methods proposed in this paper is further verified by the results.更多还原