DEVELOPMENT OF A CROWD COUNTING AND DENSITY ESTIMATION MODEL USING CONVOLUTIONAL NEURAL NETWORK

Abstract

A lot of people gather in a particular area for several purposes such as social, political or musical events. Automated crowd analysis can lead to effective and better management of such events to prevent any unwanted scene. Crowd counting remains an integral part of crowd analysis and also an active research area in the field of computer vision. One of the major problems in social gathering is overcrowding (political rallies) and crowd related disasters. Also improper planning for such events instigate these tragedies. Crowd estimation can help in reducing these calamities by providing a proper analysis of crowd especially with dense population. This study developed a Convolutional Neural Network (CNN) without fully connected layers , in which the front end was employed from a pre–trained Visual Geometry Group (VGG- 16) and the dilated convolution for the back end so as to maintain resolution of the output with the input image. Shanghai Dataset was used which is split into two parts, Part A for dense crowd and Part B for Sparse crowd. Both parts in the dataset is divided into training and testing in the ratio 70% for training and 30% for testing. The developed CNN model achieved a lower Mean Absolute Error (MSE) of 66.59 and lower Root Mean Square Error (RMSE) of 95.5 for the part A which is the dense crowd and Mean Absolute Error (MAE) of 11.0 and Root Mean Square Error (RMSE) of 14.2 for part B which is the sparse crowd compared to the previous works by Ranjan et al., 2018 in which the Mean Absolute Error (MAE) is 68.2 and Root Mean Square Error (RMSE) is 115.0 for dense crowd and the Mean Absolute Error (MAE) of 10.7 and Root Square Mean Error (RMSE) of 16.2 for sparse crowd. This shows that our method outperforms current state-of-the-art models under most of the evaluation criteria.