Research On Bridge Crack Identification Method Based On UAV Flight Path

With the rapid development of economy and society,tens of thousands of bridges have been built in China.By the end of 2020,912,800 highway bridges and more than 30,000 railway bridges have been built in China.Therefore,it is very important for the daily maintenance and repair of bridges.Among them,concrete cracks at bridge piers,abutments and main beams,as the most intuitive disease affecting bridge safety,are highly valued in bridge detection.Using traditional manual detection method and using bridge detection vehicle for crack detection has low efficiency,low safety and high cost.With the development of unmanned aerial vehicle remote sensing,digital image processing and convolution neural network technology,the method of bridge detection using unmanned aerial vehicle can break through the limitations of the above methods.However,the complete morphology of cracks can not be displayed by the crack images collected by the unmanned aerial vehicle,which restricts the follow-up research on cracks.Aiming at the problems of low efficiency and high risk of traditional bridge detection methods and unable to obtain complete crack morphology when using UAV for bridge detection,in this paper,a method of collecting bridge crack image based on UAV path and stitching it for crack identification is proposed.In the image stitching stage,by introducing SURF operator to detect the feature points of the crack image,the initial matching of the feature points is achieved based on the ratio of the nearest and the second closest distance between the feature description vectors.The RANSAC algorithm is used to eliminate some error matching pairs,and cosine constraints are added to the feature vectors to further improve the matching accuracy.A more complete crack stitching map is obtained by using a multiresolution fusion algorithm.In the crack recognition stage,the collected crack images are preprocessed by graying,smoothing filtering,histogram equalization and gradient calculation.The data set needed by convolutional neural network is made,and the U-Net network model suitable for crack segmentation is built.The optimized model is used to complete the crack identification experiment,and the experimental results are analyzed and demonstrated according to the evaluation index.In the stage of calculating crack parameters,model of UAV is selected according to image acquisition requirements and image pixel calibration is completed by pasting reference objects.The stitching method based on two images realizes the stitching of multiple crack images,and the crack segmentation experiment is carried out on the stitching image.In addition,the length,average width and maximum width of cracks are calculated by extracting the crack skeleton,and the size error of cracks in different width ranges is analyzed.The results show that the error rate of bridge crack identification using the method proposed in this paper meets the requirements of bridge maintenance and repair.