Reducing Overcrowding on Public Facilities Through Crowd Counting by Using Convolutional Neural Network

Overcrowding in public facilities often disturbs visitors' comfort. This study conducted an experiment to provide a solution for visitors who want to first see the atmosphere of a public facility through machine learning crowd counting at that place with CCTV camera sensors. Crowd counting has been utilized to reduce crowding by implementing various machine learning methods. This approach allows for precise counting of people. However, the current methods that have been used have not yet achieved perfect accuracy, due to challenges such as different light conditions and low photo resolution. This study was able to run eight experiments using the ResNet50 by differentiating the values of learning rate, number of epochs, and batch size so that it could get the best output, namely the minimum possible loss and Mean Absolute Error values which can be used as a benchmark for how accurately the machine can predict the number of visitors in the area. By using the proposed method, the best result was obtained from a learning rate of 0.0001, epochs of 20, and a batch size of 32, producing a loss value of 0.0192 and an MAE score of 0.1090. This study shows that the proposed method can accurately predict the number of people in public places, thereby this method can be implemented on CCTV cameras and the data can be integrated with mobile applications that show the number of individuals in a particular place. That way, people can check the crowd level before deciding to visit the place.