| In recent years,with the rapid development of high-speed rail,social economy,information technology and artificial intelligence,the number of people coming and going on high-speed rail stations is increasing day by day,and the area of high-speed rail passenger stations is also quite huge.Most of the high-speed railway stations at home and abroad have adopted large-scale glass curtain walls and aluminium alloy plate roofs,which can deform in bad weather such as high winds,and if foreign objects on the roofs are not dealt with in time and fall off,there are great potential risks to the public safety of the stations.Therefore,the monitoring of deformation of glass curtain walls and roofs of high speed rail stations is becoming more and more important and necessary.The traditional way of monitoring the large-scale deformation detection and identification of foreign objects on the roof of a passenger station is to use manual monitoring,which makes the efficiency and real-time monitoring very low,and a large amount of monitoring data is difficult to manage,so in order to meet these needs,it is necessary to establish an information and intelligent deformation monitoring method.The main research of this thesis is as follows.The first one is to study the method of detecting and locating the deformation of glass curtain walls and roof covers of high-speed railway passenger stations.In order to accurately detect and identify the deformation areas of glass curtain walls and roof covers,this thesis proposes to take pictures through high-definition cameras obliquely downwards,firstly using image pre-processing to streamline the complex image information and speed up the image processing speed;then following the deformation detection characteristics of the static background outside the house by employing an improved background difference method,which incorporates the weighted average method on the basis of the average background difference method to update the background area in real time,providing an effective way to cope with the interference caused by slow changes in the background and improving the accuracy of image target detection.In the deformed area localization stage,this thesis proposes to use monocular ranging for image target localization,to complete the mapping between the coordinates of the target pixels in the image and the actual coordinates,to calculate the actual coordinates of the deformed area,and to compare them with the real coordinates,and to analyse whether the calculated actual coordinates are qualified.The experimental results verify that the error between the calculated coordinates and the real coordinates is 4.33%,which meets the practical application requirements.The second is research on the identification method of foreign object intrusion detection in the roof of high speed railway passenger stations.Aiming at the problems of long-distance small target detection of foreign objects on roofs,small target pixels in images with a small proportion of total pixels and low accuracy of small target detection,the advantages and disadvantages of several algorithms for target detection based on deep learning are analyzed and an improved YOLOv3 target detection algorithm with higher speed and accuracy is proposed.The method improves the scale of the YOLOv3 network model,performs 2-fold up-sampling fusion on the third scale prediction of the YOLOv3 network to enhance shallow information and improve the accuracy of small target recognition,and gives a way to resize the anchor box using K-means++ clustering to update the YOLOv3 algorithm according to the shortcomings of the K-means algorithm that randomly initializes the cluster centroids The improved network and the YOLOv3 network were trained and tested on the data-set of this thesis respectively,and the comparison of experimental results showed that the network model used in this thesis has higher detection accuracy and efficiency.Based on the results of the algorithm research,the requirements of the deformation monitoring system and related technical methods are analyzed and selected based on the Browser/Server(B/S)distributed architecture design model,and the functions of the high speed railway passenger station building deformation monitoring system are designed and implemented,including project management,deformation monitoring,foreign object identification and system management,etc.The monitoring system reduces the manual burden,better realizes the automatic monitoring and safety management of passenger station building deformation,and improves the monitoring efficiency.In summary,this thesis combines the application scenario and needs of high-speed railway passenger stations,and takes the deformation of large-scale glass curtain walls and roof covers and the intrusion detection of foreign objects on roof covers of high-speed railway passenger stations as the research object.With the objective of improving the status quo of traditional manual monitoring,which is time-consuming and labour-intensive,at the same time,in order to reduce the workload and hardware conditions of automated deformation monitoring technology,a system is proposed to use high-definition cameras to collect image information of target objects,detect and identify the collected images through target detection technology,and develop a high speed railway passenger station building deformation monitoring system by using computer and network communication technology.The system has achieved good results in deformation detection and foreign object intrusion recognition,which has the value of promotion and application. |
