A crowd density estimation and people flow statistics method

Abstract

The invention relates to a crowd density estimation and people flow statistics method, a multi-scale fusion crowd density estimation model is used, the model is composed of a deep layer network and ashallow layer network, and the deep layer network is designed based on VGG-16. The shallow network is mainly used for learning the characteristics of a target with small pixel occupation on a picture,and the multi-scale fused crowd density estimation model extracts the characteristics of different convolutional layers of the deep network, and performs crowd density estimation by fusing the characteristics with the outputs of the deep network and the shallow network. Meanwhile, the output of the crowd density estimation model is used as the input of the people flow statistical model, so that the two models are fused together, the training speed of the neural network can be greatly increased, and the application in actual life is wider. According to the method, the accuracy of the crowd density task and the visitor flow rate same task is improved, and meanwhile, the crowd density estimation task and the visitor flow rate statistical task are completed in one model.

Description

technical field [0001] The invention relates to a method for estimating crowd density and counting people flow based on a neural network, and belongs to the technical field of video security monitoring. Background technique [0002] Crowd density estimation is the process of using machines and software to process crowd dense information, extracting its features and calculating the total number of people in the crowd density image, while people flow statistics is to extract the information between consecutive frames of pictures and calculate the number of people in a period of time. The number of people passing through a place. Crowd density estimation and people counting are often used in the field of video security surveillance. For example, video surveillance is carried out in places with a large and complex flow of people, such as railway stations and subway stations, and intelligent monitoring is carried out through group analysis to detect abnormal behaviors of groups ...

AI Technical Summary

Problems solved by technology

However, due to the large crowd density in the pictures calculated by crowd density estimation and people flow statistics, there will be many difficulties in the research, such as target occlusion, target deformation, scale transformation, etc.

For example, in a dense picture, the proportion of pixels occupied by people close to the camera is very large, but the pixels occupied by people far away from the camera are extremely small, so it is difficult for even a neural network to learn all the density information of this picture at one time.

At the same time, when the crowd density in the picture is high, the occlusion between people will be very serious, so the final estimation result is often not ideal

Images