Automatic Recognition Of H-section Steel Connector Class And Extraction Of Welding Track Based On Computer Vision

With the continuous development of artificial intelligence and Internet of Things technology,manufacturing industry is transforming and upgrading from automation to intelligence.Countries are increasing their support and investment in intelligent manufacturing industry from the aspect of policies and funds.Welding,as a basic process method in traditional manufacturing industry,plays an important role in manufacturing industry.Research on intelligent welding is of great significance to realize intelligent manufacturing.Starting from intelligent welding of H-shaped steel structural joints,the subject obtains shape and position information of H-shaped steel structural joints to be welded on welding workstation by means of computational visual technology and H-shaped steel structural joints in CAD model base.The matching identification of theoretical model of steel structural connection is carried out to obtain the type information of H-section steel connection to be welded and to extract the weld trace based on the theoretical model data in order to realize intelligent welding of Hsection steel structural connection.This paper mainly focuses on the following aspects:1.Pre-processing of H-section steel structure connection image: In view of the diversity of noise sources of H-section steel structure connection on welding station,three traditional image smoothing algorithms,mean filter,median filter and Gauss filter,are compared and experimented.The experimental results show that the noise of H-section steel connection is limited by using one smoothing algorithm alone.According to the respective characteristics of the three traditional algorithms.In this paper,a fused smoothing algorithm based on mean filter,median filter and Gauss filter is used.The experimental results show that the fused smoothing algorithm achieves better edge-preserving and noise-removing effect.In view of the uneven brightness distribution of H-shaped steel structure connection image,this paper uses Gamma correction to equalize the brightness of H-shaped steel structure connection image.The experiment shows that the image is equalized by Gamma correction.After correction,the brightness of the original image is equalized and the contrast is enhanced.2.Feature extraction of H-shaped steel structure connector: the comparative experiment between index table thinning algorithm and Zhang-Suen thinning algorithm shows that the dislocation and distortion brought by Zhang-Suen thinning algorithm are small in the process of extracting the skeleton of target image.Therefore,this paper uses Zhang-Suen method to extract the skeleton of H-shaped steel structure connector,then uses search matching method to extract the skeleton feature points,and then uses the feature points and the The distance sequence of the center points is used to construct the feature vector of the H-shaped steelstructure connector structure.The feature extraction experiment is carried out by collecting the same workpiece image from the industrial camera at different heights and angles.The experiment shows that the feature extraction method adopted has the invariance of affine transformation.The comparison experiment of the feature extraction of the different workpiece image collected from the industrial camera at the same height shows that The method of feature extraction has the distinction between classes.3.Characteristic matching of H-section steel structural joints: Firstly,the method of similarity metric feature matching based on Euclidean distance is discussed.The test shows that the method of similarity metric feature matching based on Euclidean distance has the ability to resist interference of scaling and rotating changes of workpiece image after normalization of feature vector and normalization of camera coordinate system and CAD coordinate system.The requirement of European distance similarity metric feature matching method to normalize camera coordinate system and CAD coordinate system is also discussed in this paper.Aiming at the weakness of distinguishing ability between classes when this method only uses feature vectors composed of ordered distance,the number of endpoints,crosspoints and crossover ratio are introduced to construct new feature direction.Experiments show that the improved matching method based on cosine similarity measure not only keeps the relative characteristics of the matching method based on cosine similarity measure,but also has better discrimination ability.4.System software design and implementation: Under 64-bit Windows 10 system,using Visual Studio 2012 development tool and based on MFC software framework,the type identification and welding track system of H-section steel structure connection adopted in this paper is realized.The experiment shows that the corresponding algorithm and software system adopted in this paper can match the corresponding workpiece in CAD model base and identify the corresponding theoretical model of the workpiece to be welded.According to the theoretical welding track recorded in the theoretical model and the corresponding posture transformation relationship of the workpiece to be processed,the welding track can be extracted.It is of practical value to realize intelligent welding of H-section steel structure connection.