Research On Subway Pedestrian Target Detection Method Based On Deep Learnin

As the pace of my country’s urban modernization continues to accelerate,the subway rail transit system is gradually turning to high serviceability and high efficiency,which brings great convenience to citizens’ travel and actively responds to the country’s call for "green travel,lowcarbon life".Different from traditional means of transportation,subways are mostly built underground.In the face of dangerous accidents,evacuation and rescue are relatively difficult.This requires that the internal video monitoring and analysis system and security inspection system of the subway must meet high real-time and high accuracy.In this study,a high-precision and high-real-time detection algorithm is designed using deep learning technology for pedestrian targets in subway scenes.The main research contents are as follows:1.For pedestrian targets in subway scenes,there are factors of different sizes,different degrees of occlusion,and too dark environment,which affect the accuracy of pedestrian target detection.An improved YOLOv5 s target detection algorithm is proposed to enhance the pedestrian target detection effect in subway scenes.The model adds a deep residual shrinkage network to enhance useful feature channels and weaken redundant feature channels.Using the improved hollow space pyramid pooling module,the fusion features of multi-scale and multireceptive fields can be obtained without losing image information,and the global context information of the image can be effectively captured.An improved non-maximum suppression algorithm is designed to post-process the target prediction frame and retain the optimal prediction frame of the target.Experimental results show that this algorithm can effectively improve the detection accuracy of various pedestrian targets in subway scenes,and improve the detection effect of small pedestrian targets and dense pedestrian targets to a certain extent.2.In view of the low resolution of the small pedestrian target in the subway scene,which contains less feature information,the target detector at this stage has a poor detection effect on such targets.An improved SSD algorithm is proposed to enhance the detection effect of small pedestrian targets.The model adds a pyramid feature enhancement module,and combines the multi-branch residual unit,sub-pixel convolution and feature pyramid to obtain image multiscale fusion features.The context information fusion module is used to fuse the low-level features of the image with the context features to generate an extended feature layer to detect small pedestrian targets.A dynamic positive and negative sample allocation strategy based on Anchor-free is designed to generate optimal positive samples for small pedestrian targets.Experimental results show that this algorithm can effectively improve the performance of small pedestrian target detection in subway scenes,and the effect of small pedestrian target detection with serious occlusion is more obvious.3.The output features of the convolutional layer are poor in modeling the actual target,and the shallow network cannot capture the long-distance features of the image.A target detection model based on the improved YOLOX algorithm based on Transformer is proposed to further improve the pedestrian target detection effect in subway scenes.The model uses the Swin-Transformer module instead of CNN as the model backbone network,and uses the selfattention mechanism to locate pedestrian targets in the image.A new feature fusion layer is proposed as an additional feature layer to retain sufficient feature information of pedestrian targets to improve the detection effect of small pedestrian targets.The model adds an NMA attention mechanism to the feature enhancement template to suppress unimportant image feature information.Experimental results show that this algorithm can further improve the detection effect of pedestrian targets in subway scenes.