Crowd estimation in images

• Main Author: Tang, Leroy Zi Wei
• Other Authors: Cham Tat Jen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2021
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/148105
• Institution: Nanyang Technological University

The study of crowd counting could be considered an extension of object counting, an extensively studied problem since the early days of image analysis. However, one of the many challenges identified in the early days was to overcome occlusion, which can have a huge impact on the accuracy of estimation. With Rapid technology advancement over the past few years, crowd estimation has progressed significantly with the usage of density maps. This is also the mainstream method adopted. Density maps estimation have achieved the (State Of The Art) SOTA performance for crowd estimation due to the outstanding learning abilities of deep CNNs. With the success of object counting, it has prompted the research community to look into more challenging applications such as accurate video surveillance, internal security, resource management. Entities working in related fields may benefit from this study and similarly, the farming communities may also benefit from this idea by utilising this function to count livestock. This project involves studies and works in areas such as Computer Vision, Image processing, and Pattern Recognition. In this paper, we are going to look into some of the adopted methodologies and understand the mechanics behind these methodologies that allow them to perform crowd estimation. This final year project seeks to evaluate the effectiveness of the various known methodologies. It is often challenging to accurately determine the sizes of the crowd from images. Images can be taken at varying angles and perspectives by high mounted cameras. Crowd counting or crowd estimation has evolved greatly over the past few years hence it is crucial to understand the development of the different methodologies over the years. This would fuel greater research interest in the study of crowd counting for people who follows the rise of machine learning and artificial intelligence in the recent years.