Research On Application Technology Of Remote Monitoring System For Railway Work

As the mileage of railway operations in China continues to increase,the pressure on rail security and the standards for rail maintenance becomes much higher than before,much more efficiency and safety protection is needed for railway workers.Therefore,the use of modern communication and information technology to achieve automation and intelligence of railway operation,maintenance and management is an inevitable trend.However,most railway work sections still use manual methods during operation protection.Safety incidents often occur because workers’ actions are not effectively monitored.Aiming at this problem,based on the analysis of the existing railway operation protection system,the idea of adding Human Action Recognition(HAR)in the railway work monitoring system was proposed,taking the actions of rail flaw detection workers as an example,the characteristics of the workers’ actions and the HAR method were studied,the remote monitoring system for railway work was designed and implemented.The main work of this thesis includes the following aspects:Firstly,the HAR method based on wearable sensor was studied.Including 1)action data collection;2)the methods of tilt correction,denoising,segmentation,labeling and the extraction methods of 8 types of eigenvalues;3)4 algorithms of kNN,C4.5 Decision Tree,Random Forest and SVM.Secondly,the requirements of the monitoring system were analyzed and the overall design scheme was carried out;the seven main actions of railway workers were analyzed;the functional requirements and performance requirements of the system were analyzed;the system was structured;the main three types of entities of the system were abstractly described,and the input and output data streams of each entity were analyzed;the database of the client software was designed.Thirdly,the design and implementation of the monitoring system was completed.The system mainly includes the implementation of preprocessing and feature extraction of action data,basic principle analysis of classifier-building,training and calling of models;design and implementation of network communication software of server,action monitoring device and train approach warning device;design and implementation of function modules of administrator login,user management,task creation,history query,packet parsing and processing,GIS monitoring in the Web client.Finally,the performance of the 4 algorithms and the functionality of the monitoring system were tested.The action data of 8 volunteers were collected,then preprocessed and eigenvalues-extracted.After 10-fold cross-validation by machine learning platform called Weka,the confusion matrix was obtained.The results show that the recognition effect of the four algorithms is great,and the SVM has the highest recognition rate.The SVM algorithm was used as the HAR model and the functional testing was carried out.The test results met expectation,indicating that the scheme can meet the needs of remote monitoring of the behavior of railway workers,and the existence of dangers can be discovered in time,which has certain engineering application value.