【作者】 李牧；

【导师】 赵杰；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2008， 博士

【摘要】 近年来,机器人技术己成为高技术领域内具有代表性的战略性技术之一,它使得传统的工业生产方式发生根本性的变化,对人类社会的发展产生深远的影响。随着计算机视觉和计算机硬件技术的快速发展,将视觉信息同机器人控制相结合形成视觉伺服系统,使机器人具有同外部环境进行智能交互的能力,是当今机器人发展的一个主要方向。可以预见,具有视觉的智能机器人将得到越来越广泛的应用。传统的机器人视觉伺服控制系统是基于标定技术的,整个伺服系统控制精度在很大程度上依赖于标定的精度。然而,在实际中,由于种种原因,这种基于标定的机器人视觉伺服方法受到了很大限制。无标定视觉伺服开始成为机器人视觉伺服控制领域的一个研究热点,所谓“无标定”视觉伺服是指在不预先标定摄像机和机器人参数的情况下,直接通过图像上的系统状态误差来设计控制律,驱动机器人运动,使系统误差收敛到一个容许的误差内。本文就是针对近年来在机器人视觉伺服技术领域新发展起来尚处于探索阶段,还未形成统一的理论体系的“无标定”方法展开研究的。本文对无标定视觉伺服中的目标特征点的提取方法进行了研究。采用了区域颜色和边缘信息相融合的方法,对图像进行较为稳定的分割,然后提取质心。该方法充分利用了小波多分辨率分析的特性,进行图像的边缘检测。在应用传统的小波技术对图像进行边缘检测时,需要采用阈值对模非极大值抑制后的候选边缘点进行筛选,求取边缘。目前阈值的求取是凭借人们的经验人为的设定,需要反复的试凑比较才能得出最后结果,另外,当前的单阈值自动求取方法还无法实现精确的边缘检测,这些缺陷限制了小波边缘检测技术在实际中的应用。针对这一问题,论文提出基于类内方差最小化原理自适应的求取双阈值的算法,不需要人为的设定任何系数和参数。这种自适应计算阈值的方法对各种基于梯度的边缘检测技术同样适用。采用自选定区域进行颜色分割和边缘检测信息融合的图像分割技术,实现了较为稳定的分割,较为精确的提取目标质心,必免了颜色分割存在的跳变现象以及边缘检测无法识别感兴趣目标的问题。本文提出了一种动态无标定的视觉伺服控制方法,对系统的动态残差项进行了估计。当前的无标定视觉伺服控制技术或者只能针对静态的目标,或者针对动态目标但无法摆脱系统动态残差项的影响。因此,论文基于非线性方差最小化法控制机器人跟踪运动目标,利用动态拟牛顿法估计图像雅克比矩阵,采用迭代最小二乘法提高系统的稳定性,提出对动态系统的动态残差项的估计方法,实现了机器人对运动目标的跟踪。研究了能应用于“眼在手上”视觉伺服控制结构的动态无标定的视觉伺服控制算法。当前“眼在手上”系统的无标定算法中,没有考虑到随着摄像机的运动,系统的复合雅克比矩阵会在每个时间增量时发生变化。提出了对每一时间增量时刻的图像雅克比矩阵的变化做出估计的方法。通过将非线性目标函数最小化,以视觉信息跟踪动态图像。最后,利用相关硬件组建了一套无标定视觉伺服实验系统。通过多组实验测量了系统的性能指标,并取得了预期的效果,验证了算法的有效性。更多还原

【Abstract】 Recently,Robot technology has become one of the representative strategic technology in the high-tech field,which leads the fundamentally changes in production mode of traditional industry so as to have far-reaching influences on the development of humankind soeiety.During the rapid developing of machine vision and hardware, visual servoing system is the combination between visual information and robot control,that made robot have the intelligent switching capacity with the external environment. This is one of the main direction of developing of robot. It can be forecasted that the intelligent robot with sensewill be applied more and more extensively.Traditional robot visual servoing control techonique is based on calibrated technologies,so that the control precision of the servo system depends largely on the precision of calibration. However in practiee,a variety of reasons,limit the application of the visual servoing control method based on calibrated teclnologies to a great extent.Uncalibrated visual servoing has become a hotspot in the field of robot visual servoing control.Uncalibrated visual servoing means that vision control law is designed direetly by the system state error from image plane without pre- calibrating the Parameters of camera and robot,which controls the robot to make system error converge to a permissible region.The dissertation develop the studies of robot uncalibrated visual servoing control,which is still in primary original and exploring stage in the field of robot visual serving control and doesn’t set up the uniform system info.The passage introduced the content of the research of uncalibrated visual servoing, key technologies and existing problems,introduce the feature-points extraction method and control method,moreover, the developing of uncalibrated visual servoing method and the main research content are summarized.The dissertation research on the algorithm of feature-points extraction.Fusion information of edge and area color, putting into effect of stable image segmentation,then center of interest object.The method ultilize character of wavelet multi-scale to edge detection.whatever traditional wavelet edge detection technique was adopted, threshold was needed to filter candidate edge points for edge detection. However, threshold was obtained through experience presently, the best result was received after“cut and trial method”had been used repeatly, in addition, single-threshold calculation method can not accomplish accurate edge detection by now. These shortcomings restrict application of wavelet edge detection technique in practice.Aim at this problem,algorithm of self-adaptive calculating double thresholds based on method of minimum interclass variance was proposed, and any parameter was not needed artificial setting.Fusion color of self-selection area based on segmentation information and information of edge detection algorithm, achieve better effect of segmentation,feature-points are precise extracted, jumping phenomenon of segmentation based on color is avoided,problem of can not idetify interest object by edge detection method is solved.An dynamic uncalibrated method for visual servoing lechnique is presented, The approach for estimation of mutiplicity residual is proposed Recently, most of uncalibrated visual servoing technique are only for static target and some for dynamic target but can not dismiss effect of mutiplicity residual. In this dissertation,The robot system is controlled using dynamic nonlinear least squares optimization technique to tracking moving target. Dynamic quasi-newton approach is used to estimate imagejacobian matrix. System is more stable using recursive least squaresalgorithm. The robot can track object by this algorithm.A dynamic uncalibrated algorithm for eye-in-hand visual servoing structure to track a moving target is proposed. For the change of composite image Jacobian with time is unavailable in visual servoing system now, this dissertation presents a method to estimate this change.Vision guided algorithm for tracking dynamic image is developed through minimizing nonlinear objective function.Finally, an uncalibrated visual servoing experimental system is setted.Several group of experimental data show every algorithm is of correct.更多还原