Research On Detection And Identification Of Freight Train Number In Complex Moving Scene

With the rapid growth of railway freight transportation,manual transcription of freight train number is time-consuming and laborious,and can’t not meet the needs of modern railway transportation.Considering that train number plays an important role in railway vehicle management,railway wagon tracking and train dispatching,train number identification has become a necessary work in railway management.However,different from the common vehicle license plate recognition,the recognition of railway freight train number needs to complete the data acquisition under the train moving scene,and the recognition process is more complex.Based on this,a method of detection and identification of freight train number in complex moving scene is proposed,which includes train number location,train number segmentation,single-frame train number recognition and multi-frame joint recognition,and so on.the details are as follows.Train number location: the location of train number area is the first step of train number recognition system.Because there are some non-train number interference characters on the surface of the railway wagon,and the train number is not fixed in carriage position.In order to locate the area of train number accurately,a method of locating train number from coarse to fine is proposed.Firstly,the target detection algorithm based on convolution network is used to locate the rough area of train number,based on this,it is proposed to use the train number coding features to finely locate the train number area through connected components analysis technique.Train number segmentation: to complete the location of train number,further segmentation of individual train numbers is required.The hollowed-out printing method adopted by railway train causes some of the train number character to fracture internally,which increases the difficulty of train number segmentation.Although the character of the train number is fractured internally,it is similar to the corresponding standard symbols.Based on this,a standard character template library is constructed,and the cross-correlation matching method is used to calculate the matching result of the character template in the region to be tested.In order to determine the best matching position of characters,a novel algorithm based on K-means clustering analysis is proposed to precisely extract a single train number from all the matching results.Single-frame train number recognition: for railway freight train number,the performance of the train number recognition model based on the traditional shallow network is limited.A lightweight multilayer convolutional neural network model for railway train number is proposed,which takes advantage of the powerful feature extraction capability of convolutional neural network: first of all,in order to improve the accuracy of the model recognition,a data-enhanced method is proposed to amplify the limited train number image;a lightweight neural network is used to train the model based on a large number of train number images,from which the train number character can be accurately identified.Multi-frame train number recognition: train moving image acquisition often leads to image blur,which makes the effect of train number recognition based on single frame image processing not so good.A multi-frame train number recognition model based on video spatio-temporal redundancy information is proposed: firstly,the multi-frame images after positioning are analyzed by content correlation,and the multi-frame images of the same train number are marked by segment clustering,and then a multi-frame image probability decision-maker is established,there by improving the accuracy rate of train number recognition in moving scene.Finally,the image data of the same train wagon are jointly recognized.In this work,a large number of railway freight train videos are collected on the spot.Through comparison experiments such as train number location and recognition,it is demonstrated that the proposed method has achieved good results in train number location,train number segmentation,train number recognition,and so on.The train number location and train number recognition rate are all higher than 85 %,the validity of the method in this paper is verified for the detection and recognition of railway train number in complex moving scene.