Crowd Counting in High Dense Images using Deep Convolutional Neural Network

Abstract

Crowd counting plays a significant role in analyzing the crowd behavior in high density areas. Deep learning techniques may be utilized to count the crowd from given high density images. This gives situation awareness and facilitates in imposing necessary actions to control the crowd in various scenarios when needed. In this paper a deep convolutional neural network model has been developed for crowd counting. The model has been developed using VGG16 pre-trained model and it is tuned up for crowd counting using transfer learning. The dataset used in this work is ShanghaiTech crowd dataset, that contains 482 high density crowd images. Image augmentation is applied to enlarge the dataset. The model gives a training accuracy of 83% and 79% of validation accuracy.