Research On Detection And Recognition Algorithm Of Pavement Crack In Complex Background

As an early form of pavement disease,real-time detection of cracks is of great value for pavement maintenance management and performance evaluation.The existing pavement crack detection methods are mainly based on the high-quality image,clean road surface,no shadows and interferences.The actual pavement crack detection will encounter various complicated road scenes,such as: the road has shadows,zebra crossing,oil pollution,well cover.Therefore,it is necessary to conduct in-depth research on the detection and identification of pavement cracks in complex backgrounds.Pavement cracks under complex background have the characteristics of strong background texture,uneven illumination,more interference noise and weak crack information.The key of crack detection lies in the removal of road shadow and complex interferences.In this paper,the pavement cracks with shadows,zebra crossings,oil pollution,well covers and other interferences are studied in four aspects: enhancement,de-noising,crack target extraction and characteristic parameter calculation.1)In the image enhancement processing,the MSR enhancement method is employed for the shadowed road surface crack image.From the three indicators of gray mean,standard deviation and information entropy,the enhancement effect of the global histogram equalization,local histogram equalization and MSR method are analyzed.The results show that the images processed by MSR method have higher brightness,stronger contrast,smaller information entropy,and the shadow details of the image are effectively removed.2)In the image de-noising,denoising complex interferences of road crack is important.Firstly,the characteristics of the crack image under different noise interference are analyzed.Secondly,a multi-level de-noising model is built based on the gray-scale feature,crack feature and geometric feature of the image.The de-noising effects of the mean filter,median filter,MorphoLogical filter and the filtering methods in this paper are compared and analyzed from the indexes of root mean square error(MSE),peak signal-to-noise ratio(PSNR).The results show that MSE and PSNR of image have been greatly improved,and the zebra crossing,oil pollution,and manhole cover noise on the road surface can be effectively removed by the multi-stage de-noising model.3)In crack target extraction processing,an Otsu segmentation algorithm based on L0 gradient minimization is proposed,because that Global threshold segmentation method cannot achieve threshold adaptation,Otsu threshold segmentation method produces more noise and edge segmentation method detects edges while enhancing noise.For the segmented image with noise,the segmented binary image is de-noised by the connected domain threshold method,circumscribed rectangular threshold method and background pixel threshold method.The results show that the accuracy of the crack extraction algorithm proposed by this paper is 96.47%,the recall rate is 94.08%,and the F1 value is 98.26%.The algorithm performance is higher.4)In the calculation of characteristic parameters,damage assessment and system construction,firstly,the calculation method of such parameters as length,width,area of cracks are introduced,and the pavement damage index through relevant parameters can be calculated;secondly,a complex background crack detection system based on GUI is built combined with the above research.The pavement crack detection algorithm in the complex background proposed can effectively detect cracks images with shadows,zebra crossings,oil stains and well covers,and improves the accuracy of crack target detection.It is meaningful to provide data support for maintenance management by parameters to rating the level of road damage.The system makes crack detection more complete and uniform,and is also suitable for the detection of bridge cracks.