Detecting the crowdedness of people by deep learning

• Main Author: Heng, Seng En
• Other Authors: Jiang Xudong
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Artificial intelligence; Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/154130
• Institution: Nanyang Technological University

Crowd management is a crucial aspect in the pandemic. Rapid spread of infection occurs in areas with high human concentrations. Therefore, there is a need for controlling the number of people in a vicinity. Computer vision-based AI crowd counting has been an existing area of research that could be leveraged on to enhance current crowd management efforts. In this project, various methods of Convolutional Neural Network (CNN) based crowd counting were studied, inspiring a few prototypes to be developed. Of them, the vgg19csr1 showed performance increases against baseline methods in false positive rejection and in low to medium crowds. Even though it did not attain the highest overall MAE, it fulfilled the criteria for use in context of COVID-19 Singapore. The project then ends off with the development of a functional automatic crowd monitoring system based on this model, demonstrating that AI based crowd counting has the potential to enhance COVID-19 crowd management efforts.