Crowd Analysis of Almasjid Alnabawi using convolutional neural networks of CCTV footage

Abstract

In recent months, crowd management has become more important than ever, given the spread of contagious diseases such as COVID-19. The Hajj, in Saudi Arabia, is one of the largest gatherings in the world; it happens annually and is getting bigger every year. The development of radio-frequency identification (RFID) and mobile apps has been investigated to help estimate crowd movements in and among the holy sites. However, network-based technologies require large infrastructures and are therefore very costly. In this paper, a system is proposed to use existing closed-circuit television (CCTV) to accurately visualize the movements of crowds in the Almasjid Alnabawi, also known as The Prophet's Mosque. The proposed neural network is trained with large datasets of crowd images to produce estimates of the number of pilgrims in an image. Images are then integrated to produce crowd level models throughout the building. The system has been tested on two instances and showed high performance.