Research On Orbital Foreign Object Intrusion Detection Algorithm Based On Deep Learning

With the rapid development of rail transit industry,the safety and efficiency of subway travel has gradually become the focus of our continuous attention,especially the hidden safety problems caused by high-density passenger flow in the station.Because of the large amount of computation and slow tracking speed of the traditional model,and the low tracking efficiency in the high-density occlusion environment,it can't meet the requirements of efficient,accurate and real-time tracking in the crowded crowd environment in the subway station.In view of the shortcomings of the existing algorithms,the detection and tracking algorithm based on subway passenger flow is proposed.The main research work is divided into following parts:The first part is passenger flow detection technology based on deep learning.In the normal subway station scenario,the SSD and YOLO series algorithms are analyzed.In this thesis,the basic network setting,the principle of prediction target and the training process are studied.The VOC_sub data set built by ourselves is used for training,testing and evaluation.Through the experiment comparison,the effect of the YOLO_v3 algorithm on passenger flow detection is the best.At the same time,in the case of high density and high occlusion,the conventional target detection algorithm is difficult to meet the requirements.Therefore,the CSR_Net model based on deep learning is designed to estimate the density of passenger flow.Through model training,the more accurate estimation results of passenger flow density can be obtained.The second part is passenger flow tracking technology based on deep learning.In view of the confusion between the target and the background in the traditional target tracking algorithm,which is easy to miss detection and detect errors,the single target tracking algorithm based on twin network,including Siamese-FC,Siamese-RPN and SiamMask,is introduced in this thesis,but the single target tracking needs to define the tracking passenger flow target by itself,which wastes manpower and time.In view of its shortcomings,the output of the YOLO_v3 detection algorithm is used as the input of the multi-target tracking network,and the feature extraction network setting is improved.The highly efficient deep network ResNet-50 is selected to replace AlexNet,and the deep cross-correlation calculation is used to replace convolution cross-correlation calculation to determine the tracking template.The tracking target update mechanism,IOU threshold judgment,is also configured to solvethe change of passenger flow complex problems.The comparison of the test and evaluation on the data set of this paper proves that the algorithm proposed in this paper can significantly improve the accuracy and efficiency of subway passenger flow tracking.The algorithm of passenger flow detection,density estimation and tracking based on deep learning is studied in this paper.The technical framework and algorithm flow of detecting and tracking passenger flow first and generating density map estimation are proposed,which overcomes the problems of inaccurate detection and unclear tracking existing in the traditional model,and realizes the effective monitoring of passenger flow trajectory and density distribution.The relevant research work of passenger flow monitoring has better academic promotion significance,and it has better reference value for traffic hub traffic monitoring,emergency plan,layout optimization and other applications.