Research On Crack Identification Algorithm Of Subway Tunnel Surface Image

As a kind of tunnel defeats, cracks seriously affect the safety of the tunnel. The detection of cracks in tunnels has profound impact on the tunnel’s safety. However, uneven illumination, low contrast, complex background texture and severe noise pollution are common in tunnel crack images. Detecting algorithms for traditional images like concrete surface and asphalt pavement images are not suitable for tunnel surface images. As a result an algorithm is proposed in this paper to identify the tunnel surface cracks and calculate the width of cracks. It solves all kinds of complex problems in tunnel surface images effectively and identifies the cracks accurately. In addition, the width of cracks is calculated in the pixel domain.Preprocessing algorithms like Mask Dodging and gray scale erosion are used to improve the quality of tunnel surface images. It can balance the illumination of images and enhance the contrast of cracks to make it simple for subsequent processing. Then a multi-filtering algorithm based on feature analysis is proposed. It includes filtering based on rectangular template, filtering based on connected components and linear feature filtering based on Hough transform. Large quantities of background components and texture noises are eliminated in the multi-filtering process.In order to distinguish cracks and pseudo cracks, a cracks identification algorithm based on SVM is proposed in this paper. It uses pretreatment algorithms to construct the cracks and non-cracks sample library. Then feature vectors are designed to train SVM. Cross validation and grid search method is used to select the parameters of SVM. Finally, traditional images and tunnel surface images in Beijing subway line1are tested with the algorithm. The experimental results show that the accuracy of this algorithm for traditional images can be97%and it can be81%for tunnel surface crack images.A width calculation algorithm combines binary image and gray image is proposed in this paper. Firstly, skeleton points are obtained in binary image and lots of blurs are removed in skeleton drawings. Then a two-step edge linking algorithm is proposed to link the adjacent crack fragments. Finally, a normal linear neighborhood of a crack point is picked in the corresponding position of the gray image. A principle called1-σ is proposed and the width of cracks is calculated in the pixel domain.