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 a very important role in the field of automated assembly and assembly of industrial parts. However, in complex dynamic scenarios, the grasping results are sensitive to the calibration error. Therefore, precise grasping often requires repeated calibration when the calibration parameters drift due to long-term operation. A method of picking parts based on supervised learning is proposed. The training samples are collected to obtain the image feature vector through hierarchical clustering. A positive definite kernel function is constructed and the mapping relationship between the captured state vector and the image feature vector is obtained through the support vector regression. Finally, the method can be applied to guide the online crawling. Finally, the experiment proves the effectiveness of the proposed method.