Research On Bridge Crack Measurement System Based On High-precision Calibration

Bridge crack measurement is an important part of ensuring bridge's safety.Currently,the detection of bridge cracks mainly depends on artificial work which is labor-intensive,low-precision,time-consuming,and dangerous.Computer vision measurement is the hot direction of non-contact method with advantages of high precision,high efficiency,low cost and so on.Hereby,this paper proposes a method for measuring bridge cracks accurately based on stereo vision,where the camera is calibrated non-linearly.And this method can improve the accuracy of measuring the size of bridge cracks obviously.The camera's resolution is 2560*1920,the physical focal length is 15 mm and the length of camera film is 7.52 mm.When the crack is 50 cm away from the camera,the size error of measuring crack is within 0.1mm,which approaches to the theoretical maximum pixel equivalent.Moreover,the main work of this paper is as follows:(1)Technologies of identifying bridge cracks.This paper presents the image processing methods needed to be used before identifying bridge cracks.As most images of bridge cracks are lack of illumination and have much interference,this paper will choose histogram equalization to enhance images' contrast.An adaptive median filter is used to denoise images,and OTSU binarization segmentation algorithm is selected to identify cracks.In the end,this paper utilizes Sobel operator to extract the edges of bridge cracks.Thereby,the outline of cracks will be separated from background regions.After the above process,the reliability of recognizing cracks is up to 80%.(2)Technologies of measuring bridge cracks accurately.The paper introduces how to measure distance based on binocular vision system.What is more,the principle of measuring the cracks' size has also been presented.The size parameter will be calculated based on pixel equivalent of the single camera.At the same time,semi-global stereo matching algorithm(SGM)is used to realize binocular stereoscopic vision matching.(3)Non-linear calibration technologies of binocular vision system.This paper focuses on the work of camera calibration.The significance of camera calibration is described and several major camera calibration methods are analyzed in detail.An improved non-linear calibration model is proposed in this paper.For getting the model,the quadratic polynomial will be used to fit the pixel equivalent model of the imaging plane based on corner points' pixel equivalent on the checkerboard(pixel equivalent refers to pixel's real size).Moreover,the coefficient of the fitting model is proportional to the distance from the object to the camera.Therefore,before utilizing this method to calculate cracks' sizes,in order to determine the coefficient of the fitting model,the distance from cracks to the camera has to be calculated at first.Only in this way,can the fitting equation of pixel equivalent on the imaging plane be gotten.(4)Verification of measuring bridge cracks which are based on stereo vision.The rigorous contrast experiments are set up to verify the feasibility and superiority of the proposed scheme.What is more,in order to facilitate the engineering and optimization of the measuring method,an application software for identifying and measuring bridge cracks has also been developed based on Visual Studio and OpenCV.