Subway crowd density estimation method and system based on target detection

Abstract

The invention discloses a subway crowd density estimation method and system based on target detection, and the crowd density estimation method comprises the following steps: firstly carrying out the bounding box labeling of the head of a human body in a manufactured sample library; secondly, introducing a YOLOv3 network, converting receptive fields of three scales into receptive fields of two scales by using an up-sampling operation, performing up-sampling on the outputs with different receptive fields, fusing the up-sampled outputs with outputs of the same scale, and taking the fused outputsas a final training network; and finally, determining the ROI of the detection video image frame by using a mask mode. Through human head detection, fusion detection of different scales output by a network and ROI region operation, the problem of mutual shielding among part of human bodies is solved, the problems of subway station building shielding and specular reflection are avoided, and the performance of crowd density estimation is improved.

Description

technical field [0001] The invention relates to the technical field of crowd density estimation, in particular to a method for estimating subway crowd density based on target detection. Background technique [0002] In many cities in our country, the subway has become a means of transportation for more and more people to travel, and there are more and more crowds in subway stations and the phenomenon of non-compliance with normal traffic rules. Video surveillance has become an important way to warn of dangerous events and ensure the safety of the crowd. . Therefore, real-time analysis of the crowd density of subway station video surveillance has important practical significance for urban security. [0003] Traditional detection methods of people counting rely on manual extraction of low-dimensional features (HOG, Haar features, etc.) [0004] In recent years, with the in-depth research of deep convolutional neural network (CNN) by scholars, a method for estimating crowd de...

AI Technical Summary

Problems solved by technology

For example, the application number is 201910702656.0, which discloses a machine vision-based human flow analysis method in subway public places. This method is based on the machine vision human flow analysis method in subway public places, and the scale of the feature map is added to solve the problem of pedestrians displayed in the distance due to the difference in distance from the camera. The problems of missed detection and low accuracy caused by the difference in the size of the video; the strategy of training the network from coarse to fine, optimizing the bounding box parameters (number, width and height), and controlling the balance between detection speed and accuracy

However, since the public data set it uses is a pedestrian data set, that is, it detects the "whole person", and the detection of the whole person is prone to the problem of false detection caused by occlusion

In the case of occlusion between human bodies in the subway station scene, the effect of using this method of detecting human bodies to estimate crowd density is not ideal.

Images