基于视觉引导的多类型管件识别与位姿估计研究

Research on Multi⁃type Pipe Fittings Recognition and Pose Estimation Based on Vision Guidance

针对管件在机器人抓取中的分类识别、位姿估计等问题,对随机摆放管件的模板匹配识别、立体匹配和三维位姿估计进行了研究。通过标定得到左右相机的内外参数,然后进行双目标定、矫正以及手眼标定。优化了模板匹配算法,利用亚像素级精度提取管件的边缘轮廓。选择合适的图像金字塔层数以提高匹配速度,实现多类型管件的分类识别。通过立体匹配提取管件视差图,结合标定参数和双目视觉的三角测距原理实现管件的位姿估计,并进行多组位姿提取实验。设计开发MFC和Halcon联合编程的管件识别与位姿估计系统。实验结果表明:该系统可实现管件在随机摆放情况下的实时准确分类识别与位姿估计,满足工业的实际应用需求。

Aiming at the problems of classification,recognition and pose estimation of the pipe fitting in robot grasping,the tem⁃plate matching recognition,stereo matching,and three⁃dimensional pose estimation of randomly placed pipe fittings were studied.The internal and external parameters of the left and right cameras were obtained through calibration,and then binocular calibration,correc⁃tion and hand⁃eye calibration were performed.The template matching algorithm was optimized,the edge contours of pipe fittings in tem⁃plate matching were extracted with sub⁃pixel accuracy.The appropriate number of image pyramid layers was selected to improve the matching speed and realize the classification and recognition of multiple types of pipe fittings.By using stereo matching,the disparity map of the pipe was extracted,the calibration parameters and the triangulation principle of binocular vision were combined to achieve the pose estimation of the pipe fitting,and multiple sets of pose extraction experiments were carried out.The pipe fitting identification and attitude estimation system was developed by using MFC and Halcon joint programming.The experimental results show that the sys⁃tem can be used to realize real⁃time and accurate classification and pose estimation of the pipe fittings under random placement,which meets the needs of practical applications in industry.