Intelligent Evaluation Of Road Surface Damage Based On Street View Images

In recent years,the continuous improvement of China’s urbanization construction speed has promoted the development of the road system.The safe and orderly operation of the road system and the protection of people’s safety have gradually become the main tasks of the daily work of China’s traffic department.Road surface diseases seriously affect the safe passage of highway traffic.Rapid and effective identification and extraction of road diseases have always been a major difficulty in road detection and maintenance.Past the road condition assessment using the field investigation research method,this method not only high risk and are not objective enough,and street view street view images such as baidu and tencent street view because of its wide coverage,can show a city street level of information and data sets,the cost is low,is the research of urban road pavement condition evaluation to provide new data sources.In order to make an intelligent appraisal of the condition of the road surface,this paper proposes a method system based on open network data and open source convolutional neural network to evaluate the damage condition of the road surface,aiming at the problems of the time and energy consumption of data acquisition and the difficulty of road damage information extraction.The method uses Street View images as the data source.The deep learning method is used to detect the damage information of road surface.Specific contents are as follows:(1)First of all,the crack information in the street view image is analyzed to determine the image preprocessing process: considering the environmental lighting factors,the weighted average method is used to grayscale the street view processing;Log transform is used to emphasize the low gray part of the image.In order to enhance the image contrast,the restricted adaptive histogram is used to process the image.(2)followed by the road crack information extraction,semantic segmentation method is used to choose Unet based network,in view of the road cracks in street view images of small,extremely unbalanced data characteristics,the introduction of attention mechanism construction of neural network,create images street road crack data sets,a total of 13200 copies,according to the proportion of lo,80% of the samples were used for training,and 20% of the samples were used for testing.The prediction model was obtained by training the data set established in this paper by adjusting parameters,and multiple evaluation indexes were used to evaluate the prediction results,among which the accuracy(PC),accuracy(ACC),comprehensive evaluation index(F1)and recall rate(R)reached 77% and 83%,respectively.60% and 64%,both improved compared with the basic U-NET network model,and the segmentation results were close to the recognition effect of human eyes.(3)Finally,according to the calculation of road damage index in "Highway Technical Condition Assessment Standard"(JTG H20-2018)issued by the national department,the evaluation index based on pixels was established.The street view images of expressway,main road and secondary road network in a certain area of Daxing District were selected for experiment,and the road damage status grade was evaluated on the street view images.The Arc GIS desktop tool is used to display the location information so as to realize the evaluation of the road damage condition.This research by rapidly and accurately identify the road crack,and then evaluate the damaged road condition,the method is simple in implementation and low cost,high efficiency,good reliability,put forward the method system of pavement damage condition detection provides a new research idea,for road maintenance and operation have great practical application value.