Application of Dense Crowd Detection Method Based on Lightweight Neural Network in Subway Crowd Recognition

Abstract

With the rapid development of rail transit network, scientific and effective urban public transportation management is of great significance to maintaining public order and planning transportation operations. This paper aims at the original network structure of YOLOv3, deletes the repeated data of the convolution module in the detection network, and optimizes the convolution method to design a lightweight network structure. Aiming at the problem of the complex posture and background of the crowd in the subway, the Darknet53 network with better performance is selected as the feature extraction network. At the same time, according to the actual speed requirements for subway pedestrian detection, the repeated data and parameters of the repeated convolution module in the deep network are deleted, and the original convolution method is replaced with a smaller depth convolution method. Thereby it reduces time complexity of network and improves the detection speed, and achieves the application of lightweight neural network model in subway crowd recognition.