Research On The Identification And Quantification Of Concrete Cracks Based On Computer Vision

Concrete cracks are one of the main factors affecting the durability of concrete construction.The damage caused by concrete cracks has caused huge economic losses in our country.Regular detection of concrete cracks combining with the repairing methods can greatly extend the life of the construction.Thus,concrete cracks are one of the key points of concrete research.However,the manual-detection method mainly used is timeconsuming which is difficult to reach high-precision detection standards.And pre-buried sensors which can only be buried in unhardened concrete are not flexible.Therefore,flexible,simple and high-precision crack detection technology is one of the breakthrough directions of concrete damage detection technology in the future.In this topic,the identification and quantitative analysis of concrete cracks is attempted by using machine vision technology.This method has high detection accuracy and relies on very few external equipments,which can meet the temporary detection and all-weather monitoring needs,Since it's an exploration project,it needs to be studied from the basic theory.The current problem is that in the recognition task of concrete cracks,due to the difference between the contrast of the crack image,the background complexity and the foreground complexity,the effect of crack recognition depends on the combination of image processing technology and semi-automatic parameterprocess,and is not universal,which limits the development of crack detection technology based on machine vision to some extent.In this paper,a high-precision and adaptive concrete cracks image segmentation framework is constructed based on machine vision,which is composed of convolutional neural network classification model,adaptive image enhancement method,Gaussian lowpass filter and OSTU technology.The framework which is insensitive to image contrast and foreground complexity,can eliminate complex background influence and effectively extract cracks from concrete images.Based on crack image segmentation,this paper studies the quantitative method of cracks and proposes the foreground prospect scale index to quantify the extension of cracks.The equivalent cracks model has been put forward whose width and length index can reflect the cracking situation of the cracks.The application of these indicators can describe the results of crack-identification and help analysts determine the damage-extent of concrete.