Research On Pavement Crack Detection Based On Computer Vision

Cracks are the main diseases of roads and potential threats to road safety.The detection and repair of cracks is an important measure to ensure road health.Traditional crack detection mainly relies on manual visual inspection,and the actual operation is time-consuming and laborious.Therefore,it is necessary to develop an automatic,accurate and efficient road crack detection method.In recent years,with the rapid development of information technology,the application of computer vision technology instead of traditional manual methods to complete road crack detection has received widespread attention.This article is based on computer vision technology to study road crack detection methods.In view of the small number of data sets disclosed in the current pavement crack detection research,the single type of cracks,and the high image contrast,this paper expands the scale of the data set by contrast transformation and Gaussian filtering data enhancement methods,and establishes a data containing multiple complex backgrounds and crack characteristics.set.Secondly,the application performance of four commonly used convolutional neural network models,including Alex Net,VGGNet,FCN,UNet,etc.,in pavement crack detection is analyzed and investigated,thereby revealing that the computer vision technology with convolutional neural network as the core algorithm is used to solve pavement cracks.The feasibility of detection,the general technical route,and the problems of the current common networks provide improvements and directions for the development of new detection algorithms.Based on UNet,this paper proposes a new and improved UNet network-I-UNet,and establishes a road crack detection algorithm based on I-UNet.Test analysis shows that it can effectively solve the problems of UNet segmentation inaccuracy,loss of crack edge details and obvious discontinuities.Its detection results can fuse more scale and level feature information,enrich the edge details of cracks,and achieve higher detection.Accuracy.For conventional crack images with complex backgrounds,it has a higher detection rate,higher accuracy,and stronger robustness.However,when the crack contrast is low and the crack is shallow,the I-UNet algorithm will still appear discontinuous,there is a problem of target loss,and the network training time is long and the efficiency is low.In order to solve the above-mentioned problems of the I-UNet algorithm,this paper further improves the U-Net network structure and proposes the CBAM-UNet algorithm,which is based on attention to enhance the UNet network.Fusion of shallow and deep pixel features,the introduction of convolutional attention module(CBAM)to improve detection efficiency,the use of hole convolution to expand the receptive field to reduce the loss of crack detail information,and the cross entropy loss function(cross entropy loss function)to deal with the unbalanced pixel ratio.CBAM-UNet is used for road crack detection,and compared with algorithms such as I-UNet,UNet and FCN,the results show that CBAM-UNet not only has high detection accuracy,but also has low training time and the best effect.Finally,the paper studies the feature classification and quantitative analysis methods of road cracks.Based on the two-step connected domain algorithm,the image coordinate projection algorithm and SVM classification algorithm,the crack type division is realized;the skeleton extraction method is used to calculate the crack pixel points,and the crack length,width and other quantitative information are obtained,thereby further improving the performance of crack detection.