Railway Pedestrian Intrusion Detection System Based On Thermal Imaging Video

As one of the most important modes of transportation in modern life,trains play an irreplaceable role in the field of passenger and freight transport.When pedestrians appear on the track,limited by the viewing angle and reaction speed,it is often difficult for the driver to take timely measures,resulting in a collision accident.Therefore,this paper designs a track pedestrian intrusion detection system to assist the driver to detect pedestrians on the track and issue an alarm in time.At the same time,in order to further reduce the impact of visibility on the system and achieve all-weather detection,the system designed in this paper is based on an infrared thermal imaging camera.There are three key modules in the track pedestrian intrusion detection system:pedestrian detection module,preprocessing module and track division module.In the pedestrian detection module,this paper uses YOLOv3 for pedestrian detection,and makes the following three improvements:(1)In view of the difference between infrared images and visible light images,this paper simplifies the feature extraction network of YOLOv3,and removes the existence of redundant parts;(2)In order to improve the semantic differences in the fusion of different levels of features,this paper adds an attention module to the feature fusion part of YOLOv3 to optimize the fused features;(3)In order to make the position of the prediction frame more accurately,this paper replaces the L2 loss in YOLOv3 with the CIoU loss with an added area penalty.In the preprocessing module,this paper designs a convolutional neural network with an encoder-decoder structure.The purpose of this network is to enhance the pedestrian in the infrared image and ensure that the brightness of the pedestrian is much higher than that of the background area,so that the image can be filtered according to the brightness.The area where pedestrians may exist in it is used as auxiliary information for pedestrian detection.In the track division module,this paper finds the track by edge detection and cubic function fitting,and marks the track area.The track pedestrian intrusion detection system designed in this paper takes into account the accuracy and real-time performance.It is proved by experiments that the improved YOLOv3 not only greatly improves the accuracy,but also increases the running speed by 35%.After using the preprocessing module,the accuracy of pedestrian detection is further improved.