Research On Bridge Deformation Detection Method Based On Image

With the vigorous development of China’s infrastructure construction,great achievements have been made in bridge construction projects.At the same time,the safety inspection issues during the construction and use of bridges are becoming increasingly important.The use of technical means to conduct safety inspections on bridges has always been an important research direction in the field of bridge engineering.The deformation of the bridge structure will directly affect the safety of the bridge.Aiming at the problem of structural deformation of the bridge during construction and operation,this paper uses digital image processing technology and deep learning theory to study the method of bridge structural deformation detection.The main research work of this article is as follows:1)The Dense Time Net(Dense Time Series Convolutional Neural Network)single image rain noise removal network model is proposed to solve the problem of removing rain noise from the images collected when it rains outdoors.The network is composed of multiple scale Dense Time Net Block(Dense Time Convolution Network Dense Block).The convolution down-sampling technique is used to obtain rain line feature information of different scales and the time dimension feature information that is found by time domain convolution after reducing the image dimension.Learning the mapping relationship between the rain map and the rainfree map at different scales and dimensions,speeds up the convergence of the algorithm,and at the same time makes the image feature information deep in the network.Experimental results show that compared with existing algorithms,the proposed method has better image de-raining effect and stronger image detail repair ability.2)Researched the theory of SIFT,SUFT and ORB algorithms,combined with the bridge detection application environment,for the rotation robustness of SURF algorithm,improved the generation method of SURF algorithm feature descriptor,and extracted it using principal component analysis(PCA)More valuable information in the feature descriptor.The results show that the adjusted SURF algorithm has stronger rotational robustness than the SURF algorithm,and is more suitable for the detection environment such as bridge deformation,which does not require high rotational robustness but requires high illumination and fuzzy robustness.3)Aiming at the problem of differences in the matching performance of each pair of matching points in the point set positioning process,a multi-point weighted center positioning method is proposed based on the rectangular center positioning method and the multi-point average positioning method.The main method is to assign different weight coefficients to each pair of matching points.The experimental results show that when there are matching point pairs with poor performance in the matching point set,the stability of the multi-point weighted center of gravity positioning is stronger than the multi-point average is far stronger than the rectangle Central method.4)For the long-distance large-scale scenes,the active vision calibration method and the camera self-calibration method have the problems of difficulty in operation,scene limitation,high cost,and the traditional calibration method.As the distance between the camera and the target increases,the accuracy of calibration decreases.With reference to the traditional target calibration method and active vision calibration method,a passive vision camera calibration method is proposed.Experimental results show that the scheme has the advantages of simple operation,low cost and high precision in long-distance large-scale scenes.5)Based on the previous research results,a bridge deformation detection system was realized,and real-time monitoring of bridge structure deformation was realized.