Research Of Detection Method For Tunnel Damage Based On Omni-Directional Vision Sensor

Traffic construction plays a very important leading role in the development of national economy,the highway,railway and city traffic construction,cross the rivers,valleys,mountains and across the strait or the need to build a variety of underwater tunnel engineering.Related research shows that a considerable number of tunnel disease problems,seriously affect the traffic safety and the safety of people's lives and property,so how to quickly and effectively detect the disease of tunnel has become an urgent problem.The detection equipment existing in the tunnel disease image acquisition and detection are difficult,aiming at solving the above problems,this thesis designs and implements the tunnel disease detection system,the tunnel disease detection vehicle panoramic vision sensor based on full face image acquisition for the detection of tunnel lining of tunnel lining "appearance",then the image processing panoramic image.Finally,this thesis use convolutional neural network for automatic detection of disease classification.The main research work and results of this thesis are as follows:1.A research on a tunnel defect detection vehicle for acquiring full section images of tunnel.In this thesis,the design and implementation of the tunnel disease detection system,the hardware part includes the tunnel detection vehicle,ODVS(Omni-directional vision sensor),lighting and remote computer.The detection process is as follows: the tunnel inspection car with IPC,IPC is responsible for collecting the panoramic image of tunnel lining,and then the panoramic image processing,finally will be sent after the start of the remote computer to a remote computer,through the matching software of panoramic image processing.2.Based on the digital image processing of the tunnel panoramic image of high precision automatic analysis and evaluation method.Tunnel lining concrete structure is different from common diseases,there are low contrast,spatial connectivity is poor,uneven illumination,noise of many kinds of problems,irregular distribution,so this thesis to get the panoramic image processing,including panoramic image spread,image preprocessing,image segment etc.,then,extracted damage area in the binary image processing of the optimal threshold method by four connected domain method.Finally,put forward the quantitative index of tunnel diseases and analyzes the tunnel diseases.3.A research on the recognition method of tunnel disease based on CNN(Convolutional Neural Network).There are many kinds of diseases in tunnel lining,and some diseases cannot be identified by digital image processing,so this thesis uses the convolution neural network to extract and identify the characteristics of various tunnel diseases automatically.Due to the tunnel disease data is relatively small,first of all,the data expansion of existing tunnel disease data,and then training convolutional neural network model,and finally use the training model to achieve the classification of tunnel disease identification.The thesis designs a tunnel disease detection system based on Omni-directional vision sensor,and the system of each module and the related algorithms are expressed in detail.Finally,the research contents and experimental results are summarized.The experimental results show that this method simplifies the detection device in the tunnel lining structure gets a panoramic image to a certain extent,by the end to end of the convolutional neural network to realize automatic extraction,various characteristics of tunnel disease detection and recognition,which provides effective technical support for tunnel maintenance,completion and acceptance.