Fast video crowd counting method based on time sequence relation

Abstract

The invention provides a fast video crowd counting method based on a time sequence relation. The method comprises the following steps: 1, carrying out pretreatment; constructing a lightweight neural network model; loading the video image into a lightweight neural network model; construction ofng a crowd density estimation model, loading the crowd density estimation model into a lightweight neural network model, wherein; the lightweight neural network model obtains a density map of continuous frames of the image according to the crowd density estimation model; and the lightweight neural network model converts the density map of continuous frames of the image into density features, constructs a dynamic time sequence model, loads the density features of the video image and a crowd density estimation model into the dynamic time sequence model, analyzes the dense frames of the video image through a weight calculation formula, and obtains a numerical value of a person. ; Bby building the lightweight neural network model and the dynamic time sequence model, the problems that a neural network model in a traditional crowd counting method is low in response speed and only suitable for a single picture are solved.

Description

technical field [0001] The invention belongs to the field of computer vision, and in particular relates to a fast video crowd counting method based on time series relationship. Background technique [0002] Crowd counting: the purpose is to count the number of people in the scene, and its content mainly includes density estimation and population counting. By estimating the crowd density, we can roughly know the state of the crowd as a whole, so as to make judgments on the behavior of the crowd. In order to manage crowds more safely and effectively, such as the management of short-term and high-density crowds that are prone to occur in sports fields, entertainment venues, conference centers, and shopping centers, accurate crowd flow results can be obtained by counting the number of crowds. [0003] At present, the main crowd counting methods at home and abroad mainly include two methods. One is to extract the density map of the image through the network, and then calculate t...

Images