Research Of Pavement Crack Detection Based On Level Set And Artificial Intelligence

The pavement crack detection and analysis are significant for engineering project assessment and highway maintenance management.The traditional method for crack detection is the manual measurement by inspectors,which is not only time consuming and low efficiency but also dangerous.The crack analysis relies on human experiences,which is not objective.Therefore,a convenient and reliable method for the crack detection and analysis is necessary to be proposed.In this paper,according to the characteristics of pavement crack images,it is focused on the crack detection in image processing and analysis,such as the image pre-processing,and the crack and feature extraction.The main research content and the contributions of this thesis can be summarized as follow.1.Crack pre-processing.Most of crack images acquired by a CCD camera are color level while the pre-proposed algorithms are mainly suitable for grey level images.Therefore,the color transformation will be implemented.In the crack image acquisition,due to the interferences by digital camera itself and external factors,there are shadow noises and particle noises on images,these noises are filtered out by using a geodesic shadow removal algorithm.This dissertation proposed the forward filter and backward filter to enhance the crack target and suppress noises,2.Pavement crack extraction.The results after adapting OTSU threshold as the initial area of level set evolution.Then the algorithm makes full use of the linear characteristic of pavement cracks,the level set function is improved by combining with eigenvector of the relevant Hessian matrix and the crack indication function.The improved level set function can reduce the boundary leakage during the evolution process.It can be avoided artificial setting the initial area of level set evolution by using OTSU.The new algorithm can improve the detection efficiency and receive the good experimental effect on different kinds of pavement crack images.3.Crack feature extraction and comprehensive analysis.Different kinds of pavement cracks have various characteristics,we have to discuss the common and different aspects.After crack extraction,it is essential to extract the skeleton and connect the fracture by using the proposed long and short connection method.Then the crack area and length characteristics are extracted.Finally the advantages of the proposed algorithm are verified by experiments on the amount of pavement crack images including normal pavement images and crack images.