Analysis System For Subway Tunnel Image

The subway is the focus of urban public transport development with high speed, punctuality and high passenger flow. In order to ensure the safety of the subway, periodic safety inspection must be carried on to the subway tunnel. As one of the biggest threats to the stability of the construction of a tunnel, the crack detection is of vital importance to the safety of the underground system. The traditional way of crack detection being manual, workers endure long and dangerous work every day and detect the crack with strong subjectivity. In the aim of improving the efficiency of crack detection, crack detection based on image processing has become a hot topic. However, this detection method has a very high demand on the efficiency and accuracy of the algorithm because of the huge number of tunnel images. So this thesis employ a tunnel image analyzing system with parallel processing to realize the auto-detection of the cracks with high efficiency and high accuracy.This thesis analyzes the advantages and deficiencies of the present algorithm for crack detection, and explores the image parallel processing technology. On the basis of these studies, three aspects are completed as follow:First, a crack detection algorithm suitable for tunnel image is proposed.This thesis integrate traditional pre-picture-computing programs in the analysis of tunnel pictures and come up with the pros and cons of each. According to the experiment result, the most functional program in which gray image are transferred into binary image on the basis of smoothing filtering based on gradient, Bottom-Hat and dynamic Ostu is adopt. The improved preprocessing algorithm can effectively remove the random noise, balance the light and highlight the crack information. In the process of feature extraction, four variances:the distance histogram of the sample, the linear relations of the sample, etc. are proposed to distinguish cracked and uncracked areas. And then threshold classifier and ELM classifier are designed.Second, the crack search algorithm based on percolation model is improved.The rule of initial point selection and the roundness of the percolation area are designed based on the original algorithm. By calculating roundness of percolation areas, some disturbance is removed. And then the application of probabilistic relaxation, which turns initial image into probability matrix is applied in this algorithm. Finally, the application of global threshold realizes the reconnection of cracks. As for the calculation of the length and width of cracks, the idea of differential and neighborhood extension is adopt, instead of the traditional skeleton extraction theory.Third, distributed processing for crack detection is realized.By studying HIPI-Hadoop visual system, this thesis recompose the image analysis process into three steps:image conversion, crack detection and crack information statistics. And then, designing the Map and Reduce processing functions for each step. By performance testing, conclusion that with a growing number of image, distributed computing is way more effective than single computer.