Research On Pavement Crack Detection Method Based On Multi-feature Decision Fusion

Pavement crack is the most common type of highway diseases.Timely discovering and treating pavement cracks can not only reduce economic loss,but also play a vital role in highway traffic safety.The traditional artificial detection method is time-consuming and labor-intensive,and it is hard to meet current rapid development of highways.At the same time,the artificial detection method often affects vehicle traffic,and the safety of detection environment is low.In recent years,computer hardware technology has been developing vigorously,and Al has also gained unprecedented attention.The progress of computer hardware and software makes it possible to automatically detect pavement cracks by computer.So far,many researchers at home and abroad have carried out extensive and in-depth research on this field.Due to the complexity of the real pavement environment,the influence of uncertain noise and the less explored information,the detection results of existing pavement crack detection algorithms should be improved,and the results of the quantitative performance evaluation of these methods need to be enhanced.In order to improve the precision of pavement crack detection,a preprocessing step for pavement images is proposed firstly,which is aiming at the real noise influence of uneven illumination and so on.Then,by researching from the perspectives of subblock recognition and pixel segmentation,two kinds of pavement crack detection algorithms based on multi-feature fusion are proposed.Finally,an automatic algorithm for recognizing the type of pavement cracks is designed.Generally,the main contributions of this paper are as follows:(1)A pertinent method for correcting the gray level of pavement crack images is proposed.In view of the longitudinal uneven illumination on pavement images,based on Retinex image enhancement theory,the method of longitudinal iterative gray level correction is adopted to eliminate the illumination interference.Meanwhile,anisotropic filtering is used to smooth the pavement particle background.With qualitative experimental analysis,a preprocessing step for pavement images in this paper provide a good foundation for the feature extraction of cracks in subsequent steps.(2)A multi-feature fusion based algorithm for pavement crack subblock recognition is proposed.In order to fully extract the feature information of cracks,the statistical,texture and shape features which are effective for the crack classification are selected.In view of the problem that complex noise existing on pavement images,the sparse representation classifier with better robustness is applied.Finally,based on the optimization decision theory,the sparse representation results of each feature are fused.On the pavement dataset which is collected from a real highway(HN),the detection precision and recall rate of this method achieves over 85%,which has obvious advantages compared with several representative crack subblock recognition algorithms.(3)A pavemrnt crack recognition algorithm with global and local saliency features is proposed.Based on the grayscale contrast information of pavement crack pixels,the high frequency noise is suppressed by global saliency,and the details of cracks are extracted with multi-scale local saliency.Finally,a comprehensive saliency map is constructed according to the spatial distribution characteristics of the cracks.The experimental results show that this method is superior to several representative segmentation algorithms on the accuracy and integrity for extracting cracks,its comprehensive index is better as well.(4)Based on BP neural network and geometric structure features,a recognition algorithm for pavement crack types is proposed.By analyzing the geometric structure features of the recognition results of crack subblocks,an algorithm for recognizing the type of cracks is designed.In the experiment,the average recognition precision of all types is up to 95%.