论文部分内容阅读
机器视觉在工业零件自动化抓取装配领域起着非常重要的作用.目前大多数抓取方法是基于人工干预的机器人手眼标定,然而在复杂动态场景下,抓取结果对标定误差较敏感,因此当长期作业引起标定参数漂移时,精确抓取往往需要重复标定.提出了一种基于监督学习的零件抓取方法.采集训练样本进行层次聚类得到图像特征向量,构建一种正定核函数并通过支持向量回归学习得到抓取状态向量及图像特征向量之间的映射关系,最终可应用于指导在线抓取.最后,实验证明了提出方法的有效性.“,”Machine vision plays an important role in industrial automation especially in the grasping and assembly of work pieces.Most current grasping methods are based on the hand-eye calibration.However,in the complex and dynamic scenes,the calibration error is sensitive to the scenes,and the long-term work will cause the calibration parameter to drift,so precise grasping often requires the robot to be calibrated frequently. A learning is put forward based grasping method for work pieces.This method uses hierarchical cluster to get the image feature vector,and then constructs a positive definite Kernel function.Based on the Kernel function,the support vector regression algorithm is utilized to train the mapping relationship between the image feature vector and grasping state vector.The grasping state vector can be applied to online grasping.Experimental results demonstrate the effectiveness of the proposed method.