一种面向机器人分拣的杂乱工件视觉检测识别方法

A method for visual detection and recognition of clutter workpieces for robot sorting

为了从输送带上杂乱工件中分拣出符合规格的目标工件,提出了一种基于多帧工件图像聚合分割的检测识别方法。该方法首先通过工业高精度相机获取工件图像,由改进的分水岭算法分割工件图像;然后基于工件外观形状特征,利用分类回归树(CART)对分割出来的工件图像进行类型识别;进而应用直方图反投影和核密度估计,将来自多个帧的同一个跟踪目标工件对象掩模组合成一个精细掩模,便于精确测量出工件尺寸;最后再联合机器人手眼标定参数,获取符合规格目标工件的位姿,实现机器人分拣。实验结果表明,该方法可快速从输送带上杂乱工件中准确分拣目标工件,具备良好的实用性和稳定性。

In order to sort out the target workpieces meeting the specifications from the clutter workpieces on the conveyor belt,a novel detection and recognition method based on the aggregated segmentation of multi-frame workpiece images is proposed. Firstly,the method obtains the workpiece images by an industrial high-precision camera,and uses the watershed algorithm to successfully separate clustered workpiece images. Then,basing on the shape features of the workpieces,the classification of workpiece images is performed by using classification and regression trees（ CART）. Furthermore,by applying histogram backprojection and kernel density estimation,object masks of one tracked workpiece from multiple frames are combined into a refined single one,so as to accurately measure the size of the workpieces. Finally,to achieve robot sorting,the parameters of robot hand-eye calibration are combined to obtain the pose of the target workpieces. The experimental results show that the proposed method can quickly sort target workpieces from the clutter on the conveyor belt,which indicates that the algorithm has good practicability and stability.