Research On Pavement Crack Detection Algorithm Under Complex Background

Cracks are the most common type of road pavement hazards,affecting the safe operation of highways;therefore,timely,effective detection,evaluation,and maintenance of highway pavement cracks is of utmost importance.Mainstream road crack detection methods usually assume that the road surface cracks have good imaging quality,and the road surface is clean without any shadows or interferences;however,actual road crack detecting will encounter various complicated road scenarios.The main problems are as follows:(1)Under normal circumstances,there are often public facilities such as trees,street lamps,and billboards on both sides of the road.These public facilities will cause shadows on the road surface under the sun's illumination.The shadow of the road surface will seriously damage the brightness of the cracks on the road surface.(2)In addition to shadows,the highway's road surface often contains other types of disturbances.For example,the most common interferences such as road surface oils,water stains,lane lines,rutting,leaves,etc.will be significantly affecting the detection of highway pavement cracks.(3)Although the detection and extraction algorithms of cracks have been widely studied,few scholars have paid attention to the evaluation mechanism of road pavement damage degree.Based on the above reasons,this paper conducts in-depth research on road surface shadow removal,highway pavement crack detection and extraction under complicated scenes,road surface crack parameter extraction and pavement damage degree evaluation.(1)The presence of road surface shadows not only undermines the uniformity of the brightness of pavement crack images;moreover,road surface shadows are characterized by extremely irregular shapes,large penumbra areas,and difficult to define shadow and non-shadow areas;these characteristics of road surface shadows all have brought great challenges to the smooth detection of cracks on the road surface.To solve this problem,this paper proposes a road surface shadow elimination algorithm based on adaptive brightness elevation model-SGRSR.The specific steps of the algorithm are as follows:First,the morphological dilation operation and Gaussian smoothing filter are used to eliminate the effect of pavement cracks and road surface texture on the subsequent shadow region partitioning;then,the maximum entropy threshold segmentation is used to solve the Gaussian smoothed road image shadow region.The partition thresholds of the non-shaded area and the non-shaded area are used to realize the self-adaptive determination of the dividing threshold.Finally,the elimination of the road surface shadow is achieved based on the improved brightness contour division model and the brightness compensation method.(2)As the actual pavement crack detection process will encounter a variety of complex road scenarios,and the mainstream highway pavement crack detection and extraction algorithms cannot be a good solution to the detection and extraction of pavement cracks in complex scenarios.Considering the great success of deep learning in image segmentation recently,how to apply convolutional neural network to the detection and extraction process of highway pavement cracks is of high application value.This paper applies the convolutional neural network to the detection and extraction process of pavement cracks.It constructs and implements a deep neural network structure that combines a convolutional neural network and a deconvolution layer neural network,directly on the road.The crack image on the road surface predicts the semantic category to which it belongs at the pixel level;then,using a specific color,the area of the pavement crack is marked;finally,based on this color feature,the image of the marked road crack image is segmented and extracted cracks on the pavement.(3)After the detection and extraction of pavement cracks,how to evaluate the degree of damage of the pavement based on the extracted pavement cracks,and how to quantify the extracted pavement cracks,so as to provide a reference to the road surface damage indicators for highway maintenance.To solve this problem,the quantification of road surface cracks and the evaluation algorithm of pavement damage degree were studied.The core idea of this algorithm is:First,build a road pavement crack image data set with a certain scale through manual acquisition and image data set amplification algorithm;then,based on AlexNet network,build a six-category network of classifying cracks such as crack-like crack,block crack,longitudinal crack,transverse cracks,network cracks,and pit damage;and finally,based on the detection results of pavement cracks and classification results of pavement cracks,quantitatively describe pavement cracks,and according to the classification results and quantitative description of cracks,the degree of damage was evaluated in order to provide a reference for road maintenance managers.