A crowd analysis method based on an attention mechanism and a deformable convolution neural network

Abstract

The invention provides a crowd analysis method based on an attention mechanism and a deformable convolution neural network. The method comprises the following steps that in a training stage, firstly,an attention map generator (AMG) is trained through a crowd image and a background image, and the trained attention map generator model is used as a component trained by an auxiliary density map generator (DME); in the testing phase, only the trained density map generator is used to generate the corresponding density map for the input crowd image. The invention combines the attention mechanism togenerate an attention diagram to detect the area of a crowd and to reflect the crowding degree of the crowd area to a certain extent. The attention map is used as a priori knowledge of the crowd to train the deformable convolution neural network, so that the network can overcome the uneven distribution of crowds in the crowd scene, environmental noise and other problems, and generate accurate crowd density map.

Description

technical field [0001] The invention relates to a crowd analysis method based on an attention mechanism and a deformable convolutional neural network. Background technique [0002] With the widespread use of surveillance cameras and people's increasing concern for public safety, crowd analysis has attracted the attention of many researchers in recent years. In order to meet the needs of practical applications, crowd analysis has also developed from simple crowd counting to displaying crowd distribution characteristics through density maps. Crowd density maps contain more spatial distribution information of crowds than simple crowd numbers, which can help us make correct decisions in crowded high-risk environments and prevent accidents such as riots and stampedes. [0003] At present, the methods applied to crowd scene analysis are mainly divided into three categories: detection counting, regression counting and density map analysis. Detection and counting implements crowd ...

AI Technical Summary

Problems solved by technology

However, there are two problems with this method: first, this method needs to scan the entire image through a moving window to calculate the number of people, which is very time-consuming; second, the detection method is blocked in a crowded environment, and the environment is noisy. and other effects, the counting result is inaccurate

However, these methods still suffer from a decrease in accuracy when faced with problems such as uneven crowd distribution in crowd scenes and environmental noise.

Images