Investigation Of Crack Recognition Algorithms Based On Analysis Of 3D Contour Features On Asphalt Pavement Surface

With the gradual completion of highway construction in China,the main mission of related departments gradually changes from pavement design and construction to maintenance,among which the detection and repair of pavement cracks are vital components of maintenance tasks.The extension and development of cracks will have a negative impact on the service life of pavement and the comfort of driving.Therefore,it is of great significance for transportation engineers to accurately obtain pavement cracking data and timely repair them.Related researchers in various countries have developed highly automated non-contact techniques to efficiently detect pavement cracks so as to reduce or get rid of the dependence on artificial labor,of which the research of digital imaging technology is the mainstream.In recent years,for the sake of solving the problem of 2D image quality being susceptible to many factors such as illumination,shadows,road marks,ruts and oil stains,some scholars have developed 3D imaging technology based on structural light.Different from 2D imaging principle,this new technology is mainly based on triangulation principle,using cameras to capture the contour of the pavement surface and imaging through 3D reconstruction eventually,so as to better solve the shortcomings of 2D image being vulnerable to environmental interference.However,due to the presence of low contrast of cracks,uneven background,changeable crack patterns,complex pavement texture and other objective factors,accurately extracting cracks from 3D pavement images is still a huge challenge.So,this research topic will concentrate on 3D pavement crack detection technology,and aiming to cope with current algorithms' problem for low accuracy and poor reliability,four crack recognition algorithms are proposed in this paper based on morphological analysis of 3D crack contour characteristics on asphalt pavement surface.At the same time,a crack connection algorithm based on probabilistic relaxation and a denoising algorithm based on morphological analysis of the fitted external ellipse of cracks are designed in this paper to increase the accuracy of crack detection.The research contents are as follows:(1)crack detection based on product of height difference: this method utilizes the feature of high-low-high degree of crack profiles being greater than that of texture region and the strong symmetry of crack profiles,to design a product operator of height difference,which can effectively enlarge the difference between cracks and non-crack region or texture region and eventually improve the identification accuracy.After examining two textured 3D asphalt pavement images: smooth and rough,the result shows that this algorithm can achieve 93.73% of comprehensive accuracy,78.08% of comprehensive recall rate and 82.58% of F value.(2)crack detection based on match of cosine function: by tuning the designed cosine function to match the image contour or profiles,the proposed algorithm can precisely extract cracks from background according to the fact that the match response of crack profiles is greater than that of non-crack area.The test result shows that the proposed algorithm can effectively identify the cracks in small width and low depth(low contrast)that average algorithms cannot well handle with,and can reach 91.89% of accuracy,90.67% of recall rate and 90.61% of F value.(3)crack detection based on multi-feature test: firstly,this method analyzed three main features of cracks in 3D asphalt pavement images: tilt-level,Gaussian-distribution and edge-gradient.Then,the corresponding feature tests are devised to realize the precise extraction of cracks according to the property of the three features.In addition,the proposed algorithm can optimize the parameters in the gaussian distribution test based on the roughness of the pavement surface,making the recognition precision meet the requirements of practical project.The results indicate that this algorithm can obtain 88.71% of accuracy,94.55% of recall rate and 90.98% of F value.(4)crack detection based on model of profiling centroid in local coordinates: this method analyzed the centroid distribution of two geometric figures which are determined by the contour on two sides of vertical axis and horizontal axis in local coordinate system.Among all the possible six types of contour or profiles,the centroid distribution of the two figures for central points and edge points in crack profiles is highly different from that of other profiles.Therefore,the threshold segmentation methods can be used to extract central points and edge points of crack profiles.The experimental result shows that this algorithm,which refers to the property of section centroid in material mechanics,considering the whole shape and ignoring the secondary or unimportant information,can effectively suppress random noise,and has a high accuracy rate(94.57%),recall rate(84.48%)and F value(88.23%).(5)Crack connection based on probabilistic relaxation: probabilistic relaxation is a pattern recognition method considering neighboring objects and gradually eliminating the ambiguity through iteration so as to find local optimal solution and the focal object is eventually labeled by a certain symbol.Some scholars have made some special designs for this method to suppress and eliminate the random noise in 2D images of cement concrete pavement.This paper will improve it based on the special designs to connect cracks.The results show that the improved algorithm can effectively realize the aim.(6)Denoising based on morphological analysis of the fitted external ellipse of cracks: this method analyzed the eccentricity of the fitted external ellipse of crack fundamental element.It is found that the short cracks are long and narrow strips,and its eccentricity is larger.While,the block noise tends to be round and its eccentricity is relatively smaller.The experimental results show that by judging the range of eccentricity of the fitted external ellipse of cracks within a certain scope of area,it can effectively filter out block noise and retain short cracks at the same time,thus improving the accuracy of crack detection.The presented methods and ideas in this thesis provide a new perspective and important reference value for the research on how to improve the accuracy of crack detection based on 3D asphalt pavement images.