The Research On Online Detection Method Of Tractor Parts Based On Machine Vision

Machine vision online detection technology is the use of industrial cameras instead of human eyes,intelligent detection algorithms instead of experience and traditional computing methods of the emerging detection technology.Because of its high efficiency,low cost and non-contact characteristics,the online detection method of machine vision has great application prospect in the fields of identification and location of parts,size measurement,surface quality inspection,assembly assistance and so on.How to realize the online detection of machine vision efficiently and reasonably has become a research hotspot in recent years.Taking traction machine parts as the research object,this paper studies an online detection method based on machine vision,the main research contents and achievements are as follows:(1)This paper introduces the online detection method and research status of Machine vision,and summarizes the development trend of online detection of machine vision on the basis of analyzing the demand of online inspection of traction machine parts.(2)According to the characteristics of traction machine parts,the basic scheme of online detection of traction machine parts is worked out,and the two problems of intelligent recognition and image mosaic in the process of online detection are summarized.This paper introduces the basic theory of image preprocessing,feature description,machine learning and feature point matching and image fusion in the process of intelligent recognition.(3)Aiming at the problem of feature recognition of traction machine parts without fixture positioning,an intelligent recognition method based on supervised machine learning is proposed.This paper studies an image preprocessing method suitable for traction machine parts and realizes the acquisition of the basic geometric parameters of feature holes in combination with Hough transform,proposes a feature description method based on the geometric position relation of feature center,and then introduces how to realize intelligent recognition by using machine learning method in combination with this method.Finally,a method of error correction based on feature uniqueness is proposed to correct the identification error.(4)Study an image mosaic method based on KAZE algorithm aiming at the local image mosaic problem of the traction machine parts.Firstly,the matching area of the image to be stitched is estimated by using the motion parameters of the three-axis moving load table.After analyzing and comparing the surface characteristics of traction machine parts based on SIFT,SURF and KAZE algorithm,the KAZE algorithm is selected as the basic basis for finding the matching feature points of adjacent images to be stitched together,and the registration method is modified and supplemented,and finally the optimal suture method is used to realize image mosaic.(5)Build a test platform for verifying the online detection method of the traction machine parts,and give the test results of the above theory and method.In the experiment of intelligent recognition,the comparison results of image preprocessing and machine learning algorithm are given,and a complete case from feature matrix generation to error correction is given.In the experiment of image mosaic,the comparison diagram of feature point matching results of several methods and the comparison diagram of image mosaic results are given.