Research On The Railway Level Crossing Object Detection And Recognition Algorithm Based On Deep Learning

At present,due to the complexity of topography and capital needs,many cities in China still have some original railway crossings.In such original railway crossings,how to ensure the safety of trains,pedestrians and vehiclesis an urgent problem for railway departments to solve.The traditional railway level crossing security relies on human resources for video monitoring.Now,with the continuous development of deep learning,this monitoring method has gradually become intelligent.This paper mainly studies the pedestrian and vehicle detection and identification methods of railway level crossings,combined with deep learning technology.The specific research contents are as follows:（1）First of all,For the freight railway level crossings in the city,This paper builds on the traditional dataset,PASCAL VOC2007,Part of the data is expanded with web crawler technology,Supplement the van categories needed for my own experiments,Written as the dataset MY,A total of 10,557 data sets were expanded;second,Image denoising and image enhancement for some data containing noise;last,For the image color distortion problem of the traditional dark channel algorithm,A dark channel deogging algorithm for k-means clustering segmentation combined with grayscale histogram is proposed,Improved the video image quality in the railway level crossing scene.（2）The yolo v3 and the yolo v4 target detection and recognition models are deeply studied and improved for my study scenarios.Firstly,soft non-maximum suppression is used to solve the target occlusion problem.Secondly,a yolo v4 network model based on high-level semantic embedding module is proposed to realize the reuse fusion of high and low level feature information.Finally,the improved algorithm is analyzed to verify the effectiveness of the proposed algorithm.（3）Based on the improved yolo v4 model presented in this paper,a prototype system for pedestrian and vehicle detection and identification at flat railway crossings is designed and realized.The system can detect and identify human and vehicle targets for real-time video and imported video,and realize the fog processing of video images,and can also voice remind the detected people and cars,and assist the traffic management of relevant railway departments.This paper combines deep learning technology with Py Qt5 graphic program framework to solve the detection and identification of people and vehicles in the railway level intersection scene,and has voice alarm function to assist the traffic management of the railway level intersection,which is of great significance to ensure the safe operation of trains.