Research On Complex Interference Removal Method Of Pavement Crack Image

Pavement image crack detection is of great significance to scientific pavement maintenance decision,which has gained widespread attention,as well as a great number of research achievement.However,in practical application,the pavement image to be inspected usually has two kinds of complex interference.One is the interference caused by pavement traffic marking and the other is the interference of light and shadow caused by natural light or artificial light,which limits the effect of most existing detection methods.As for the removal of that complex interference,this paper carries out the research on new methods to remove the interference of road traffic marking as well as light and shadow,which is of significance to promote wider application of pavement image crack detection both theoretically and practically.The main task of this thesis is as follows:1.A Generative adversarial network(GAN)based method of removing traffic markings on is proposed.Essentially,the removal of traffic markings is regarded as a pavement image repair problem.Firstly,Otsu threshold segmentation is used to obtain the area where traffic markings is located.Then,a generative adversarial network model based on texture and structure hybrid attention module is proposed,and a training set composed of randomly generated mask images and lossless pavement images is defined.The trained generative adversarial network model is used to carry out generative repair to the segmented area.The experimental results show that the proposed method can remove the interference of traffic markings and at the same time retain the existing crack information.2.A Robust Principal Component Analysis(RPCA)based method of removing vertical stripe light and shadow is proposed.The multi-functional road detection vehicle developed by the research group owns a special strong reinforcement light system,which eliminated the light and shadow interference caused by natural light in the process of collecting pavement images,but also inevitably introduced the vertical stripe light and shadow interference.At first,the observation matrix is constructed,and then the matrix is decomposed by the RPCA algorithm to remove the light and shadow from the original pavement image.The experimental results show that the quality of pavement image has been significantly improved.3.A U-Net based pavement image crack detection method is proposed,and a comparative experiment on the effect of this method before and after interference removal is carried out systematically.The experiment results show that without improving the U-Net model and its training algorithm,the road traffic marking removal and vertical stripe light and shadow removal,as a pretreatment means,can significantly improve the effect of the U-Net method,which shows the effectiveness and necessity of the proposed interference removal method,thus it has practical application value.