Research On Pavement Crack Detection Based On Computer Vision

With the rapid development of roads and railways,the increase of various types of vehicles and mileage,and the speeding up of railways have exerted tremendous pressure on pavements.Whether it is road pavement or railway pavement,pavement cracks can be seen everywhere,traffic accidents caused by cracked pavement happen from time to time.Research on pavement crack detection method is of great significance to pavement maintenance and traffic safety.In recent years,most of the pavement crack detection methods are based on image features and machine learning.The methods based on image features include edge detection,histogram analysis and morphological methods.Although such methods can achieve rapid detection,more thresholds need to be set manually and different thresholds need to be set for different data sets or different images in the same data set,and heavily relies on the selection of thresholds.In the past few years,a large number of scholars use supervised learning to detect pavement crack,such as support vector machine,neural network,structured random forest,etc.These methods generally need to mark the features first and then train through the model,can greatly improve the accuracy of crack detection,but the disadvantage is that the detection performance depends on the data set selection and slow learning,so the algorithms are difficult to be generalized.In fact,the crack detection needs both performance and efficiency.Therefore,this paper mainly studies the unsupervised method of pavement crack detection,and improves the crack detection efficiency under the premise of ensuring the performance.The main work of this paper are:(1)An adaptive unsupervised crack detection method is proposed.In order to overcome the problems of supervised learning adaptability and poor promotion ability,the block-based unsupervised crack detection method is studied,which has the advantages of fast speed,good adaptability and easy extension to different types of lines.However,there are also problems such as mistaken detection and missing detection.To solve this problem,an adaptive block crack detection method based on context information is proposed for missing detection cracks.Aiming at the mistaken detection cracks,a method for analyzing crack geometrical structure based on connected areas is proposed.At the same time,the data set is expanded,which not only applies the algorithm to the road pavement,but also validates the validity of the algorithm on the railway pavement dataset.(2)The algorithm of representation of pavement crack characteristics is realized,the classification of pavement cracks and the basis for determining the severity are given,and the verification is made in the test sets of road and railway pavement.(3)The pavement crack detection system is designed and implemented.The optimized crack detection algorithm,crack type and severity classification algorithm are applied to the system and tested on different data sets.