Research On Method Of Welding Path Planning Based On Binocular Machine Vision

The traditional teaching welding robots are less intelligent,having no spatial perception.With the continuous development of machine vision technology,the introduction of vision system enables robots to have spatial perception and the ability of autonomous welding.Taking the vision system of welding robot as the research object,this thesis studies the 3D reconstruction process of welding path in detail.Furthermore,great efforts have been made to study and improve the method of welding seam marking,image defogging and target detection algorithm based on the vision system of welding robot.After the analysis of the three most typical extraction methods,this paper proposes a method to use feature points to continuously mark the characteristics of welding seams based on the existing extraction methods of feature marks of welding seams.In addition,the OpenCV library was invoked on the Visual Studio 2012 platform to verify the ORB and SURF feature point marking algorithms.The Forstner,Harris and Sojka feature point marking algorithms were verified on the Halcon platform.The author finds that Forstner algorithm has obvious advantages in the precision of feature marking and noise control.The image quality of robot welding system is greatly affected by external environmental factors,especially the light intensity of working environment.When the ambient light intensity is too weak,the visual system may not be able to mark the characteristics of the weld;For this reason,a defogging algorithm was introduced to enhance the image quality of the vision system.On the basis of systematic analysis of existing fogging algorithms,this paper selected 8 fogging algorithms under the Matlab platform for comparative analysis,covering He Kaiming algorithm,2 kinds of histogram processing algorithms,4 kinds of Retinex algorithms,homomorphic filtering and fogging algorithm.Through comparison,it is found that MSRCR algorithm is the most suitable object of this paper.As for the problem of structural feature interference in the extraction process of welding path,such as similar structural features in the welding target,the feature point marking algorithm cannot distinguish these structural features as welds.In this paper,the method of creating ROI area on the welding target is used to filter the weld structure noise.In order to improve the location accuracy of the ROI region,a target detection platform based on FasterRCNN algorithm was built on Matlab to realize the dynamic creation of the ROI region of the weld.