Rail Transiffic Safety Video Detection Technology And System Implementation Based On Deep Learing

Due to the rapid development of China’s economy and the increasing expansion of urban scale,the construction of rail transit is speeding up,and its importance is also increasing.As a result,the scale and complexity of China’s rail transit continue to rise,and the accident rate is also increasing year by year.Therefore,it is necessary to apply video image processing technology to urban rail transit.In recent years,due to the explosive growth of data volume and computer computing power,deep learning technology has developed rapidly and has been widely used in different fields,among which image processing is one of the most widely used research fields.This paper mainly studies the application of object detection technology based on deep learning in urban rail transit video monitoring,builds a complete system framework,uses deep learning technology to extract the important information of monitoring screen,and judges and records the dangerous behavior.This paper presents a network structure and algorithm based on deep learning model which can effectively improve the detection performance of the camera front end.By comprehensive comparing a variety of target detection model,including YOLOv4 because of its high efficiency and high precision characteristics,determined for detection model of the system is mainly used for improving the camera front detection performance,reduce the detection rate of false positives,designed a kind of alarm filtering network structure based on target detection,the integration of the front drive,information interaction,filtering algorithm,and other functions,The identification server with target detection function is used to carry out secondary identification processing on the alarm picture information sent by the front end of the camera,specifically identify the target category of the intrusion object,and effectively filter the false positives triggered by the front end of the camera due to the passable target.Since the recognition end only needs to detect and recognize front-end alarm pictures,one server of the recognition end can guarantee real-time intelligent monitoring of hundreds of cameras,making full use of the computing power of the server and the camera itself,greatly reducing the cost of hardware installation and improving the freedom of intelligent monitoring.In addition,the model is improved by adding a preprocessing layer of image enhancement,which improves the detection accuracy of the model for night images.This paper studies the statistical function of human traffic in the actual monitoring scene based on the deep learning model.According to the specific characteristics of the actual rail transit application scene,a large number of real images with large population density are collected and produced,and a target detection model suitable for the rail transit scene based on deep learning is given after retraining.It effectively avoids the problems of the traditional morphological frame difference method,which is greatly affected by the change of scene illumination and the extremely serious occlusion between people in the traffic environment,which leads to the reduction of detection accuracy,and effectively improves the robustness of the system and the accuracy of target recognition.In order to ensure the real-time and accuracy of the human traffic statistics,The YOLOV4-TINY model is selected as the basic model of the human traffic statistics after comparative test.This paper presents an arc flash detection algorithm of pantograph based on dispersion analysis in rail transit scene.By observing a failure in the bow net arc flash duration is very short,the characteristics of high brightness,arc flash detection algorithm based on state tracking,and for the algorithm can adapt to more complex traffic scenes,avoid leaves reflector lamp phenomenon caused by the false positives,by analyzing the record the size of the flash pixels discrete degree,for the further improvement arc flash detection algorithm based on discrete degree analysis,The accuracy of detection is greatly improved and the false alarm rate is reduced.This paper designs and implements an intelligent monitoring and detection system for urban rail transit safety based on deep learning.Build the hardware framework including image data acquisition module,GPU image processing module,data communication module and client human-computer interaction module,as well as the software framework including image recognition processing,violation judgment,alarm information transmission and alarm.The hardware construction and software programming test of the whole system are realized,and the detection and recording of flash detection,regional intrusion and illegal cross-border abnormal behavior under video monitoring are successfully completed,with good test effect.