Estimate the size of crowd by KERAS and CNN

Abstract

Currently, estimate the size of crowd is very necessary due to the complicated situation of the epidemic. There are many methods of crowd estimation currently proposed such as the method based on the HOG algorithm combined with SVM, the regression-based approach (using Fourier and SIFT analysis), the method using the MCNN, but the mean absolute error (MAE) and mean square error (MSE) are still high. In this report, I have estimated the number of people in the crowd using CSRnet which is a convolutional neural network with ten hidden layers, and the data after each estimate is stored on thingspeak.com