Research On Key Issues Of Tunnel Lining Crack Detection

Cracks in the highway tunnel lining will affect the safety of the tunnel and require regular maintenance.Using machine vision to replace manual inspection can greatly improve efficiency.Although a lot of relevant research has been carried out,it has not been applied to the annual tunnel maintenance in China.The difficulty encountered is that the collected tunnel lining images mainly have issues such as dark light,uneven illumination,and large background noise.This makes the cracks have different gray scales,contrasts,and widths,and overlaps with noises,resulting in low-contrast areas that are easy to miss detection,and noise-like areas are easy to misdetect,resulting in low accuracy of crack detection.In order to solve the above-mentioned issues,this paper has carried out research on several key issues of tunnel lining crack detection.In this paper,a tunnel crack detection vehicle designed and developed by our research team is used to build a database of lining images captured at the Tangling tunnel in Liaoning Province.This paper takes the tunnel concrete lining crack as the detection object.Based on the in-depth study of its characteristics,it adopts the idea of combining micro and macro to design and optimize the crack detection and splicing algorithm.Through theoretical research and experimental analysis,the algorithm ideas and the selection of key parameters have been improved,and the tunnel lining crack detection technology has been formed.The related algorithm software is compiled independently,the image acquisition device is designed,and the tunnel lining crack detection system is developed.The detection system built in this paper has good detection accuracy and stability in the practical application process of Tangling tunnel lining crack detection in Liaoning Province,which provides theoretical analysis and technical support for the further application and promotion of the technology.The main research work and contributions are as follows:(1)Aiming at the issue that low-contrast cracks are easy to miss,most current line detection theory is based on the theoretical framework of image and kernel function convolution.For different detection objects,it is limited by the size and dimension of the convolution kernel.This paper proposes a crack detection algorithm based on roof ridge features.Since the roof ridge appears as extreme points in the image,a new idea of converting the line detection in two-dimensional space into the extreme point detection in one-dimensional space is proposed,without image processing,the roof ridge features are directly extracted to complete the line detection.This method has universal applicability to all line detections with roof features,and is not affected by light and shade,width and contrast,and achieves zero missed detection,but there is a certain amount of false detections.Then,the similarities and differences of roof features between cracks and the noises are analyzed,and the probability model belonging to the cracks is established through the histogram of roof height.The experiment shows that the noise removal rate is 92.8%.Finally,combine the mathematical morphology method to connect the break points to extract the cracks.In addition,the influence of the concave-convex relationship of adjacent roof ridge features on crack detection was discovered,and an optimization algorithm for fitting the illuminated surface was proposed.(2)Aiming at the issue that the cracks at low contrast and the noise at high contrast caused by uneven illumination are easy to confuse,this paper analyzes the projection curve of the image under the framework of Gaussian multi-scale sub-pixel edge detection theory,and finds the influence relationship between projection gray and contrast,and clarifies that this influence complements and perfects the existing theoretical framework.A dynamic partition Gaussian crack detection algorithm based on the distribution of projection curves is proposed,a dynamic partition criterion is constructed,and the influence of the three parameters of gray,contrast and width on crack detection is discussed.And through the recall rate,accuracy rate and time efficiency curve to estimate the optimal parameters in the algorithm,the experimental results show that the recall rate of crack detection can reach more than 98%.In addition,to solve the issue of crack breakpoints,a broken line connection algorithm based on sector template search is proposed,and combined with mathematical morphology to extract cracks,the accuracy rate can be increased by more than 80%.This algorithm provides a new idea for detecting lines with different widths in unevenly illuminated images.(3)Aiming at the issue of not being able to directly splice the end,caused by continuous shooting and missed shooting of consecutively collected adjacent images,for the first time,the spatial topological relationship RCC theory is combined with image point features,and the existing RCC theory is improved in describing the spatial topological relationship between points and line targets.This paper proposes a long-crack connectivity feature point recognition algorithm based on RCC,which can automatically identify whether the cracks in two adjacent images belong to the same long crack and whether there is overlap or gap phenomenon.Then,for the overlap phenomenon,a method based on cosine similarity is proposed.The crack registration algorithm of performance measurement can automatically detect the overlapping area and complete the high-precision splicing after registration;for the gap phenomenon,the geometric calibration method is used to complement the gap and the cracks are connected end to end to complete the low-precision splicing.The experimental results show that the accuracy of splicing is 93.2%,and the efficiency is increased by 43%compared with the feature detection method.This method provides a new idea for judging whether there are continuous shots and missed shots in the continuously collected gallery,and how to stitch together continuous lines in multiple images.Through the above work,it is possible to achieve robust crack detection under low-contrast conditions in a single image,as well as the extraction of cracks with different brightness,light distribution,different types,and different widths.While,it can realize the identification and stitching of long cracks in multiple images.