Research On Detection And Recognition For Foreign Body Along The Railway Based On Deep Learning

Railway is the backbone of the transportation in our country.The business philosophy of railway is transportation safety and efficient development.Railway is the foundation of the railway transportation.One of the prime reasons for railway accidents is that the protection is not strong enough and foreign body comes into the restricted area frequently.It is important to ensure the safety of railway transportation by monitoring the foreign bodies along the railway.Under the new situation of high speed rail operation,it is difficult to ensure the effectiveness of the foreign body recognition,which is time-consuming and laborious.So it is urgent to establish a set of automatic foreign body monitoring platform which is suitable for railway operation.The main research topic is detection and recognition for foreign body along the railway based on deep learning.The purpose is to replace the eye with machine vision.The function of the system is automatic detection and recognition of foreign bodies that may cause the safety of train operation.It gives a real-time alarm when there are security risks.The aim of the system is to help railway station make the decision to remove the foreign body and repair the train track and provide the necessary clues for the further design of rail protection facilities and management of train operation.The main research work is as follows:(1)The paper introduces the research status of foreign theory of railway safety monitoring and deep learning and analyzes the significance of automatic monitoring platform of foreign railway safety.Then it is of necessity to apply the machine vision and deep learning to railway safety monitoring system.(2)The railway video acquisition system based on Raspberry Pi is designed.The main module of video capture is Raspberry Pi 3B equipped with a special camera.The peripheral equipments include GPS,3G,SD and alarm modules.Finally the construction of hardware acquisition system is completed.The hardware is controlled by Python language in the Raspbian Jessie system.Then the system achieves real-time video acquisition of railway.(3)Image preprocessing techniques such as image gray,smooth filtering and image enhancement are studied.Through the analysis of the characteristics of foreign body image,an improved Gauss filter method is designed which is better than other methods.Using histogram equalization of image contrast enhancement after enhanced processing,experimental result shows that the two methods are beneficial to retain the image edge information and improve the measured image identification.(4)According to the characteristics of the dynamic change of railway video background,a moving object detection method based on Gauss mixture model and the three frame difference method is designed.In the foreign body recognition part of railway,this paper analyzes the advantages and disadvantages of the method of image classification.After studying the theory of deep learning,the paper improves the algorithm of CNN network as the recognition and classification of railway foreign bodies.Then the scheme is implemented by setting up the training set,pre training and testing the foreign body image.The experimental result shows that the proposed method can improve the recognition rate effectively.(5)According to the problems existing in railway foreign body safety monitoring,this paper presents a solution to apply deep learning algorithm to foreign body monitoring platform.The foreign body monitoring platform based on B/S structure is developed.And the depth learning is integrated into the monitoring platform to realize the recognition and classification of foreign bodies.The experimental result shows that the detection and recognition rate are higher than that of the traditional method.Finally,the paper summarizes the work and points out the next research direction.