Crowd density estimation method based on multi-scale convolutional neural network

Abstract

The invention discloses a crowd density estimation method based on a multi-scale convolutional neural network, and in order to improve the accuracy of crowd density estimation in the field of video monitoring, the invention provides a network structure based on the multi-scale convolutional neural network, which can accurately predict a crowd density map in a scene. According to the method, feature fusion of different receptive field information is carried out by using the cavity convolution and the original convolution, and different levels of semantic information of the feature map under different resolutions are fused, so that the crowd density map with higher quality is generated. An experiment is carried out on a current popular ShanghaiTech data set, a UCF_CC_50 data set and a WorldExpo '10 data set, and an average absolute error (MAE) and a mean square error (MSE) are used as evaluation criteria. The result shows that compared with the previous method, the network model has theadvantages that the MAE value and the MSE value are reduced, and the crowd density estimation accuracy is improved.

Description

technical field [0001] The invention relates to a method for estimating crowd density based on a multi-scale convolutional neural network, which belongs to the technical field of computer vision. Background technique [0002] With the improvement of the quality of human life, large-scale group activities such as festivals, concerts and sports events are becoming more and more frequent. In recent years, group emergencies caused by dense crowds have become the focus of society. Crowd density estimation, as an important method of crowd control and management, is an important research topic in the field of intelligent monitoring today. It can not only count the crowd in the current scene to help staff manage effectively, but also can pass Predict certain abnormal behaviors of the crowd, plan for emergencies, and strengthen the safety of public places. [0003] Traditional research methods can be mainly divided into two types, one is based on detection, and the other is based on...

AI Technical Summary

Problems solved by technology

However, there are still many deficiencies in the above method, which are mainly reflected in the following aspects: due to the wide structure of the multi-column convolutional neural network, more time is wasted in training; in the process of crowd density estimation, a density classifier is needed, In the process of using the classifier, there will be a large amount of calculation, and a large part of the parameters in the network are used for the density classifier, while only a small part of the parameters used for density map estimation, so additional workload

Images