Research On Classification And Detection Method Of Pavement Crack Based On Machine Vision

In recent years,with the vigorous development of highway construction in China,the scale of roadway management and maintenance has also grow n rapidly.Among them,roads,especially the highway,are prone to cracks due to long-term operation,resulting in traffic safety hazards.At present,only the manual recognition detection method or the single image processing detection method can not meet the development speed of the road surface crack detection,and there is a certain error.How to accurately identify and detect the crack target existing in the road surface in real time and repair it now is an urgent problem to be solved.The current classification of pavement images is based primarily on feature extraction and the use of machine learning classifiers.However,there are many kinds of surface features that can be extracted.How to choose effective and suitable features,so as to achieve a good distinction between cracks and background in the road image requires comparison and combination of different features.For the detection and location of crack images after classification,a series of image processing procedures are mainly used,which involves different threshold parameters for different shapes and depths,which need to manually set different threshold parameters for segmentation,screening and extraction,how to further improve the image processing process,or using the most widely used deep learning model for crack image detection,then improving its versatility,practicability and accuracy more effectively.Specifically,the main contributions of this paper are as follows:Firstly,the feature parameters of HOG+LBP combination are used to classify and identify road cracks.Secondly,by using SIFT feature so as to compensate for the excessive size dimension of HOG+LBP feature vector,and use the SURF feature to further improve the time complexity of the SIFT feature algorithm.Finally,through the classification recognition effect of the SVM classifier training model of the machine learning,to compare the accuracy of the three different feature extraction methods.On the image detection processing method designed for crack targets,the improved iterative threshold segmentation algorithm is used to segment the crack target,and the crack connected domain target screening algorithm and convex hull-based fracture crack splicing algorithm are designed to improve the accuracy of the existing detection and location method.Citing the YOLOV3 model of the convolutional neural network in deep learning,which further improves the accuracy of the recognition and location of pavement crack targets.