Crack Detection And Width Calculation Based On Computer Vision

Crack detection and width calculation of highway pavement images are important steps for the transportation system to maintain highway quality.Traditional manual detection and statistical methods are accompanied by subjectivity and consume too much human resources.With the development of computer vision,intelligent transportation systems have emerged.Crack detection and width calculation based on computer vision has become an important means to highway quality detection and evaluation.This paper makes in-depth exploration and research from two aspects: crack detection and width calculation.In order to improve the accuracy and efficiency of crack detection and width calculation,the following research work is carried out in this paper。Firstly,this article summarizes the research background and significance of crack detection and width calculation,as well as the research history and current situation.Analyzing the advantages and disadvantages of the existing crack detection and width calculation methods,and summarize their possible problems in practical applications.In response to these problems,this paper proposes our method,and sequentially explains the related technologies used in this method,including deep learning,generative adversarial network,conditional generative adversarial network,cyclic generative adversarial network,U-Net network,principal component analysis algorithm,robust principal component analysis algorithm and random sample consensus algorithm.Secondly,this paper proposes an unsupervised crack detection algorithm based on mapping transformation.In practical applications,crack detection faces the following challenges:(1)There are fewer datasets labeled with cracks,and the number of labeled images at the pixel level in the public dataset is very small;(2)The boundaries of the cracks are not clear,and there are errors in the labeling of the datasets;.(3)The gap between the highway environment over time is too large.Existing supervised crack detection algorithms are sensitive to the environment and are not universal.To solve these problems,this paper proposes an unsupervised crack detection method.This method uses an improved U-Net generation network and a fully connected discriminative network to play against each other.In order to reduce the sample space of the generated network,this paper introduces a cyclic consistent network and a cyclic consistent loss function to obtain a mapping model of the original crack image to the binary crack image.The experimental results show that the crack detection effect of this method is more accurate than the existing methods based on machine learning and some supervised methods,and the qualitative vision has better results.Finally,a crack width calculation method based on cascade principal component analysis is proposed.The crack width calculation method proposed in this paper is a pixel-level calculation method.This method mainly uses the cascade principal component analysis algorithm to calculate the principal axis of the crack,and then counts the number of pixels of the crack along the direction perpendicular to the principal axis of the crack.The cascade principal component analysis algorithm is a layered algorithm for calculating the principal axis of cracks.First,the principal axis of the crack is calculated using the principal component analysis algorithm,and then the accuracy of the principal axis is verified using a random sample consensus algorithm.For some major shapes with complex cracks and inaccurate calculations,the principal axis of the crack is recalculated using a robust principal component analysis algorithm.This not only improves the accuracy of the algorithm,but also improves the efficiency of the algorithm.Experimental results show that calculating crack width based on cascade principal component analysis has a higher accuracy.