Research And Application Of Weld Extraction Method For Plate Based On Line Structure Optical Visual Sensing

Visual technology based on weld feature point identification is an important part of automatic and intelligent welding.It is of great significance to simplify the process of weld identification and find a weld identification scheme with stronger anti-interference ability and wider versatility for the application of intelligent welding technology in the market.Based on active vision,this paper studies the calibration of visual device and the image in welding process,aiming to explore a method of extracting the 3D coordinates of image feature points.Firstly,the visual device is designed.including the selection of industrial camera and the selection of lens.through the study of the measurement principle of structural light vision device,the position relationship between industrial camera and structural light emitter is determined.650 nm filter is used in industrial camera to cooperate with650 nm of structural light to solve the problem of strong light interference during welding and install baffle plate to reduce splash interference.Then the visual device is calibrated.Zhang Zhengyou calibration method is used in the calibration of industrial cameras.In the aspect of calibration of structural light,a simple calibration scheme based on collinear three-point perspective principle proposed by comparing the traditional calibration methods.In this paper,the reliability of the scheme is confirmed by adding complex background information to the calibration image,and its accuracy is verified in the subsequent experiments.This scheme has certain application value in complex environment.Then the paper studied the image feature extraction of weld seam.for image preprocessing,the ROI region is extracted based on the gray distribution characteristics of the image,and the image is de-noised by median filter convolution operation.On image processing,GMM algorithm is used to extract the highlight part of the image(the way of prediction classification),and the highlight area in the image is trimmed by de-islanding and morphological operation,and then the algorithm is improved for the deficiency of the Zhang-Suen bone scaffold extraction method(poor robustness in weld contour extraction,partial distortion of the extracted diagonal line)and by interrupting the graph.After removing distortion and least square fitting,the contour points of V weld are obtained by the area filtering operation.Finally,the paper carries on the experimental analysis.the calibration accuracy of the camera device is verified.at the same time,the stability and real-time performance of the feature points extraction of the weld image are tested.the 3d coordinates of the weld feature points in the camera coordinate system are extracted and the error analysis is carried out.The experiment shows that the error of the calibration scheme is less than 0.07 mm,the precision is high,the speed of the extraction algorithm of weld feature point is 80-95 ms/ frame,which meets the real-time requirement of automatic welding,and the feature point extraction is more stable under the interference of strong light impact construction,and the horizontal error of weld feature point is 0.12～0.45 mm,high The degree error is 0.35～0.78 mm,with high accuracy.