Research On Station Passenger Flow Detection Based On Convolutional Neural Network

Because of its high safety and high timeliness,high-speed rail has become the preferred travel mode for almost everyone.With the increasing popularity of high-speed rail and the continuous application of new technologies independently developed in China,the passenger flow of each station is also increasing year by year.The purpose of research and experiment in this paper is to detect and count the passenger flow in the station scene based on deep learning and target tracking algorithm.Convolutional neural network（CNN）and visual object tracking are also widely concerned and studied in the field of computer vision in recent years.YOLOX target detection algorithm absorbs the latest achievements of target detection in recent years on the basis of YOLO series.However,the core purpose of visual tracking usually focuses on the following two points: one is to effectively estimate the action trajectory of the object in the subsequent video sequence;the other is to clarify the activity state of the object in the subsequent video sequence,and analyze the deep content of object semantics based on the collected information data.Based on the field of computer vision and guided by the latest research achievement in the direction of target detection,namely YOLOX,focuses on the multi-target tracking section in the direction of target tracking.At the same time,in the research process,it is based on the multi-target tracking strategy represented by simple online and real-time Deep SORT,And embed it into the actual tracking task.This thesis makes a detailed analysis of the current two-stage target tracking algorithm,which is based on the research of the traditional target tracking algorithm and the statistical algorithm of Deep SORT and YOLOX,and then makes a detailed analysis of the current two-stage target tracking algorithm.After expounding the basic principles and processes of YOLOX target detection algorithm and Deep SORT target tracking algorithm,the last step of the whole system,namely the pedestrian flow counting algorithm,is introduced.Finally,the pedestrian flow statistical algorithm is tested on MOT16 data set,and the video sampling is carried out in Shenyang north station,and the pedestrian flow statistical algorithm is applied to the actual station scene.In this thesis,the characteristics of target detection algorithm are fully considered,the YOLOX model is modified,and a YOLOX-AM target detection model based on CBAM attention mechanism is proposed.Firstly,the following two modes are mixed and applied in the pedestrian flow counting and statistics system.One is the YOLOX-AM target detection algorithm,and the other is the Deep SORT tracker,Due to the embedding of the above two modes,the tracking algorithm can be successfully applied to the pedestrian multi-target tracking dataset MOT16,and12 evaluation indexes are calculated.At the same time,the pedestrian flow counting function is realized on the MOT16 data set and applied to the station scene.Through experimental verification and analysis,the passenger flow statistical algorithm based on YOLOX-AM and Deep SORT has achieved good results in the practical application of station environment,which reflects the good robustness of the system,has practical application value,and can be extended to application scenarios such as airports,schools,shopping malls and etc.