Research On Key Technology For Tunnel Monitoring Of Inspection Robot Based On Image Mosaic

Tunnel engineering plays an important role in traffic operation.Accurate and efficient disease detection and traffic safety monitoring technology can improve the level of maintenance and management tunnel engineering,and ensure the operation of tunnel traffic to be safe and stable.Machine vision technology in ITS(intelligent transport system)can sufficiently guarantee the traffic safety,improve the operation efficiency and level of management the traffic system.The machine vision is widerly used in the maintenance management and non-destrutive testing of tunnels and some other structures since it has advantages of visual,high precision and easy saving data,etc,.However there are still many problems to be solved because of the partiularity of tunnel engineering,such as small application range,low automation level,high energy consumption,cost and so on.The panoramic system provides all the information of the visual that the view is given in the environment through image mosaic technology,and then rely on these visual information to make more accurate analysis,judgment and decision-making of the environment.Its technical advantages are distinct.In this paper,the tunnel monitoring technology of inspection robot based on image mosaicing is studied,in order to achieve accurate and efficient detection of cracks,leakage water disease and traffic safety monitoring in the tunnel engineering,and the reliability is verified by experiments in the actual environment.The specific research work is as follows:(1)In order to solve the problems that scene is limited and the calibration process is complicated even time-consuming when using image mosaic of camera calibration parameters,a cylinder image mosaic method based on fast camera calibration in multi scenes is proposed.First of all,the feature of extraction of high accuracy by the feature of chessboard calibration board is used to make it respectively located in the overlapping field of two adjacent images,the image sequence is preprocessed by corner extraction,precision and matching of chessboard in order to accurately and quickly solve the registration parameters between the spliced images.then the registration parameters obtained by calibration are used to quickly splice the image,and the image is projected through the cylinder In order to maintain the visual consistency of the image,and adopt multi-band fusion to retain the details of the image;finally,the whole system is built on the lowpower embedded platform,which can complete the process of fast calibration and splicing based on calibration parameters in multi scenes.The experimental results show that this method can complete camera calibration accurately and quickly in indoor and tunnel scenes,and the process of image mosaic is time-consuming.At the same time,it can ensure high mosaic accuracy and good imaging effect,and has strong robustness.(2)The real tunnel environment is more complex,because of poor light conditions and strong environmental interference,it is easy to come out error signals that affect the detection accuracy.Therefore,the traditional machine vision and deep learning design algorithms are used to improve the accuracy of tunnel crack,leakage water detection and object recognition,so as to improve the performance of the monitoring system in this study.For crack detection,a adaptive threshold algorithm based on Canny is proposed,This algorithm strengthens the effect of image denoising,uses the adaptive gradient threshold method to suppress non maximum value,and adds the rule of feature correlation to make judge,which can fully extract the crack features more and improve the accuracy of crack detection;for leakage detection,a multi neighbor-region dynamic image segmentation algorithm is proposed,the object is segmented by the method of calculating the maximum entropy threshold of the two-dimensional gray histogram of the image,and the dynamic area connection is carried out by the m-adjacency method,which can sufficiently carry out the statistics of the characteristic information of the water leakage;for object recognition,through the improvement of the traditional Tiny YOLOv3 framework and the combination of sliding window program,using the improved neural network in panoramic image to perform calendar detection,which makes full use of the existing large number of monocular vision data sets,rather than the less panoramic image data sets,so as to improve the accuracy of object recognition.Finally,the performance of detection and recognition of the above algorithm is verified by related data sets.(3)The panoramic image mosaic platform is designed and built,and the inspection robot with the platform inspects on the specific track in the tunnel,the proposed method is tested and verified in the actual scene.Based on the flushbonading operating system and ROS software architecture,the image mosaic of inspection robot,the objectt detection and recognition is designed and developed,and the transmission of information and data is carried out through a complete network system,so as to achieve the automatic monitoring of cracks,leakage water and traffic safety in tunnel engineering.Experiments are carried out in different scenes,in order to verify the feasibility of the designed image mosaic and visual monitoring method.