Research Of Three-dimensional Reconstruction Of Welding Pool Surface Based On Binocular Vision System

Welding process is widely applied in modern manufacturing industry,which determines the quality of modern manufacturing products.It is an indispensable factor to promote the rapid development of modern manufacturing industry.The traditional control of welding quality mainly depends on the design of welding parameters before welding and the inspection after welding.It cannot be adjusted in real time according to the actual welding process,and cannot meet the needs of modern manufacturing industry.To improve the weld quality of manual welding,welders correct welding parameters or the position of welding gun based on the information of welding pool surface acquired through their eyes and professional experience accumulated in long term working.It means that welding pool surface includes important visual information for the skilled welders to control welding process.The development of intelligent welding control can be promoted by 3D reconstruction of welding pool surface.Binocular stereo vision method has the advantages of simple equipment structure,high measurement efficiency and high reconstruction accuracy.It is one of the most researched vision sensor detection technologies in the field of noncontact real time quality control.Therefore,it is of great practical significance to reconstruct the3D shape of welding pool surface under the arc with binocular stereo vision.In this study,images of the welding pool under the arc were captured by the binocular stereo vision system.The improved SURF-BRISK-KAZE algorithm was proposed to detect,extract and match the welding pool feature points.The improved RANSAC algorithm was designed to eliminate the mismatched feature point pairs.The world coordinates of the feature points were calculated based on the preset coordinate system,and then the 3D shape of welding pool surface was reconstructed.The reliability of the algorithm is verified by the tracer particle experiment,which provides a strong guarantee for promoting the development of intelligent welding.An improved SURF-BRISK-KAZE feature points detection and matching algorithm is proposed.Aiming at the problem that only few feature points can be detected by a single method,SURF,BRISK and KAZE feature points were extracted by the improved algorithm.Thus,abundant feature points of welding pool were obtained.Because only the brightness information of welding pool image was used in the process of feature points detection,the color information of feature points and their neighboring pixels was introduced in the process of feature points description.The variance and mean value with color information are added to the traditional SURF feature points descriptor to form 70-dimensional improved feature points descriptor.Experimental results show that the improved SURF-BRISK-KAZE algorithm can get the most feature matching pairs,and the algorithm has strong robustness.The improved RANSAC algorithm was proposed to eliminate the mismatching of feature points.Aiming at the shortcomings of the traditional algorithm that the number of iterations is too huge to meet the real-time detection,a data preprocessing model based on slope and Euclidean distance and a pre-inspection model were proposed to improve the efficiency of the traditional algorithm.The improved algorithm avoids a series of unnecessary verification iterative process.The improved RANSAC algorithm was compared with the traditional algorithm by using standard database images and welding pool images.The experimental results show that the efficiency percentage of the improved algorithm is at least 160%higher than that of the traditional algorithm.Based on the preset coordinate system,the relationship between the pixel coordinates of the welding pool images and the actual world coordinates of the welding pool was constructed,and then the welding pool points cloud was obtained.The points cloud was processed by the LOWES S fitting algorithm,and then the 3D shape of the welding pool surface under the arc was reconstructed.The reliability of feature points matching algotithm,mismatching elimination algorithm and reconstruction results were tested by tracer particles experiment.