Pavement Crack Extraction Based On Path Morphology

With the constant development of our country's economy and the increasingly perfection of infrastructure such as roads,the total number of roads in China has reached the top of the world.Road maintenance has become a top priority for the road transportation sector.Pavement cracks are one of the most common diseases in roads,which seriously affect the service life of roads and the safety of vehicles.Rapid and accurate identification of cracks in roads is conducive to the timely maintenance of roads by the relevant traffic department,prolongs the service life of the roads,and reduces the driving safety of the vehicles.The traditional manual detection method is slow and easy to be interfered by external factors,and the detection accuracy and time are not guaranteed.The automatic crack detection based on image processing is the core task of the research.In this paper,the following studies have been carried out to address the enhancement of crack images and the extraction of cracks:(1)Crack image denoising based on path morphology and median filter:analysis all the possible noise types comprehensively considering the data acquisition process,the road's own conditions and the surrounding environment.The noise in crack images are classified by two categories according to the image characteristics of the noise:Locally small non-linear noise and large area noise.For the noise classification results,the path morphology and the median filtering method are respectively adopted to eliminate the complex background interference noise while protecting the crack's own linear structure.The combined denoising method by path morphology and median filtering provides excellent data for subsequent high-accuracy extraction of cracks.(2)Research on crack extraction based on Gabor filter banks:Analysing the causes of cracks formation and summarizing the characteristics of cracks in pavement images from the following aspects including grayscale,shape and texture.Two-dimensional Gabor filter bank is proposed to extract cracks with different widths,lengths,and directions.According to the analysis of crack width,grayscale,structure and other characteristics,a reasonable Gabor filter bank parameters are configured to perform experiments on cracks with simple structures,cracks with complex structures,and crack images in complex backgrounds.Experiments show that the method can effectively extract cracks in the pavement image.(3)Fast calculation of two-dimensional Gabor filter:Gabor filter is the algorithm which closest to the surface neurons of human visual field.It is widely used in many fields such as edge extraction and linear target recognition.However,the Gabor filter has a large amount of computation,and the computational time of a multi-scale and multi-directional Gabor filter bank is very slow.Therefore,this paper studies the two-dimensional Gabor filter principle and decomposes the two-dimensional Gabor convolution kernel into two one-dimensional convolution kernels approximately.At the same time,implements the CUDA calculation of Gabor filter based on the multi-thread image processing method of GPU.