Research On Weld Recognition And Tracking System Based On Binocular Vision

As an important technology in the manufacturing field,welding is widely used in the industrial fields of shipbuilding,aerospace,automobile manufacturing and other industrial fields with its advantages of reliability,stability,low cost and suitable for the connection of a variety of metal materials.Since the introduction of "Industry 4.0",the manufacturing industry has put forward higher requirements for welding automation,precision and intelligence.Most of the current welding systems are non-linear,and the control method of programming and teaching lacks the detection of assembly errors and thermal deformation of weldments during the welding process,resulting in poor welding quality.To solve these problems,a welding seam recognition and tracking system based on binocular vision is designed,and conducts welding seam tracking experiments on typical V-shaped weld grooves.First,the hardware composition of the system is introduced,including the gantry three-axis mobile platform,the welding torch,the binocular vision sensor and the PC104+ bus control box.A binocular vision sensor is designed,and the hardware selection,structural parameters and effective field of view of the sensor are theoretically analyzed.Secondly,the calibration algorithm of binocular vision system is designed.The binocular camera were calibrated by using the MATLAB toolbox.The matching and3 D reconstruction of the feature points were realized by the Bouguet stereo correction algorithm,and the measurement accuracy of the binocular camera was analyzed.When the internal parameters of the camera are known,a pose estimation algorithm based on a single feature point is used to simplify the hand-eye calibration model and obtain the hand-eye calibration result.Thirdly,the welding seam image processing algorithm is designed,which mainly includes ROI extraction,image morphological filtering,image binarization,and single pixel centerline extraction.The welding seam fitting algorithm is improved.On the basis of distance filtering,the accuracy of welding seam fitting is improved by subpixel corner detection and pixel interpolation.Finally,the two-dimensional information of the weld feature points is converted to the three-dimensional position information under the base coordinate system,and the weld feature points are fitted to complete the weld track detection and weld tracking experiments.Experiments show that the welding seam trajectory detection error is within,the welding seam transition is uniform,and the welding quality is good,which verifies the reliability of the binocular vision system design and image processing algorithm.