面向低纹理工件的识别与位姿估计方法研究

Research on identification and pose estimation methods for low texture workpiece

针对低纹理工件识别和位姿估计中,目标特征难以提取,导致识别和位姿估计较为困难的问题,提出一种将深度学习和轮廓重建相结合的方法。首先,将双目摄像机采集的目标图像经过深度学习处理得到工件类别和所在区域;然后,进行直线提取,并对左右图像的同名目标区域中的直线集合进行匹配,进而重建出工件的轮廓点云;最后,和CAD辅助工具生成的模板点云进行配准得到工件的位姿。实验结果表明:该方法不仅能对低纹理的工件进行识别,而且能够快速精确地计算其6D位姿。

Aiming at the problem that features of target are difficult to be extracted during the identification and pose estimation of low-texture workpiece, which leads to the difficulty of identification and pose estimation, a method combines deep learning and contour reconstruction is proposed.Firstly, target image acquired by binocular camera through deep learning processing to obtain workpiece categories and area, and then, straight line extraction is carried out, and straight line collection in the same name target area of the left and right images is matched and reconstruct the contour point cloud of workpiece.Finaly, template point cloud registration generated by CAD auxiliary tool is carried out to obtain workpiece position.The experimental results show that this method can not only identify the low-texture workpiece, but also calculate its 6 D pose quickly and accurately.