Research And Implementation Of Image Defogging And Anti-jitter Algorithm In Deformation Monitoring Based On Machine Vision

The deformation of engineering structures will pose a serious threat to safety,so it is very important for the construction and operation of the project to accurately and timely monitor the deformation form and then take corresponding measures.In engineering deformation monitoring,the global navigation satellite system(GNSS)measurement method has poor monitoring accuracy due to signal occlusion in some scenes,and artificial measurement method and synthetic aperture radar(SAR)measurement method are also difficult to be applied due to lack of real-time performance.Aiming at these problems,this paper introduces the method of machine vision for deformation monitoring.This method has been applied in engineering due to its advantages of low cost and automation.However,in the process of monitoring,haze will degrade the image quality and affect the monitoring accuracy.However,camera shake will lead to image offset or even blur,which will bring coarse error and affect the monitoring effect.In order to solve these problems,this paper first introduces the scale invariant feature transformation algorithm to extract the displacement of the monitoring target.Then,corresponding solutions to haze and camera shake are proposed respectively.The specific achievements are as follows:(1)Aiming at the problem that it is difficult to apply the monitoring method of machine vision in foggy environment,a monitoring method combining Dark Channel Prior(DCP)is introduced.Firstly,the haze concentration of the image is calculated by using the haze concentration discriminator to determine whether it needs to be defogged.For images that need to be defogged,DCP algorithm is used for defogging processing to restore their texture details.Then the displacement extraction algorithm is used to extract the displacement of the object and judge its deformation.Taking slope monitoring as an example,the comparative experiment results show that the algorithm performs well,with an error of only 0.25 pixels in fog scenarios,while the Dehaze Net model with the best performance among other algorithms has an error of 1.33 pixels.In the supplementary experiment,the average error of the algorithm is 0.4 pixel,and the running time is 67 seconds,which proves that the algorithm basically meets the requirements of deformation monitoring in accuracy and real-time.(2)Aiming at the problem of image blur caused by camera shake,a fuzzy judgment method based on Mean Subtracted Contrast Normalized(MSCN)variance is introduced.Based on the MSCN coefficient,the variance mean of the coefficient matrix is extracted as the criterion for fuzzy image judgment.Taking bridge pier monitoring as an example,a comparative experiment is carried out.The results show that the success rate of fuzzy image recognition of this method is about 85%,and the success rate of Sobel operator,which is the best among other algorithms,is only about 68%.Experimental results show that this algorithm has better recognition success rate than other algorithms and is more reliable.(3)Aiming at the problem of image offset caused by camera shake,a jitter correction algorithm based on stable reference is introduced.In this algorithm,the displacement of the stable reference object is extracted as the jitter shift,and the coarse displacement of the monitored object is corrected.The experiment was carried out on a railway bridge in Fengyang,and the results show that the displacement error of the image is large,up to 12 pixels,under the influence of camera shake.However,the displacement accuracy of the image can reach less than 0.8 pixels,and the operation time is 16 seconds.The results show that the algorithm has good performance in precision and real time,and can effectively eliminate the effect of jitter.(4)According to the algorithm introduced above,the deformation monitoring image processing software based on machine vision is designed and implemented.The software realizes the core functions proposed in this paper,such as displacement extraction,image defogging,fuzzy recognition and jitter correction.Some common image processing functions are also added,which can basically meet the needs of image data processing in actual engineering.