Similar Defect Recognition Of Cold Rolled Rust-free Strip Surface Based On Binocular Stereo Vision

The surface defect detection of cold-rolled stainless strip steel mainly uses machine vision technology for on-line detection and identification,but cold-rolled strip steel surface inevitably exists such as scratches,inclusions and other gray and form similar defects,using conventional machine vision detection,the accuracy of recognition is low.Two-eye stereo vision technology,using the twoeye camera at the same time to shoot the target of the parallax map to achieve three-dimensional reconstruction,with simple system structure,low cost,high efficiency,accurate and other advantages,can be used to identify objects with depth characteristics.In view of the similar defects that cannot be recognized by conventional visual methods of cold-rolled rust-free strip steel surface,a threedimensional identification method based on two-eye stereo vision is proposed to detect and identify similar defects with deep differences.The main research work is as follows:First of all,a two-eye stereo vision rust-free strip steel surface detection system is constructed.Two black-and-white cameras with resolution of resolution,bar light sources and reference laser single light source to build a two-eye stereovision detection system,wherein the angle between the two-eye camera is 30 degrees,and placed at the same height vertical lying down in the center of the striped steel sample;To solve the problem of stereo matching caused by the lack of texture on the surface of cold-rolled strip steel.Based on the principle of binocular stereovision,the structured two-eye stereovision detection system uses MATLAB for camera calibration,and the average error of calibration is 0.35.Secondly,the hough transform method of the homomorphic filtering of defect-splitting of the steel surface of the reflective rust-free strip steel is proposed.Homomorphic filtering divides the image into two parts,irradiance strength and reflection strength,which greatly eliminates the effects of reflection due to uneven lighting and rust-free steel surface by suppressing the irradiance and increasing the reflective component,and the hough line transform finds the position of the linear defect slot of the cold-rolled rust-free strip steel surface by counting the peak of the edge image in the Hof transformation space,which provides the basis for the target segmentation.The experimental results show that the line defect area of the cold-rolled rust-free strip surface image can be detected accurately by this method.Finally,a random forest and support vector machine detection and identification model of a two-eye stereo-matching parallax map are established.Using the constructed two-eye stereo vision platform to collect the defect image of the surface of the cold-rolled rust-free strip steel sample,the two-dimensional defect target segmentation database is established,which contains 71 mixed defect images and 81 scratch defect pictures;The defect database is divided into training sets and test sets by 5:1,in which the training set contains 57 mixed defect pictures,scratch defect pictures 64,the test set contains 14 mixed defect pictures,scratch images 17,respectively,using random forest and support vector machine two classifiers for training identification.The experimental results show that the detection accuracy of the two-eye stereo-matching parallax chart is the highest,up to 94.In the detection and identification based on the two-dimensional defect database,the accuracy of the random forest detector is the highest,up to 90.