Crowd estimation in images

• Main Author: Pui, Chin See
• Other Authors: Cham Tat Jen
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2023
• Subjects: Engineering::Computer science and engineering::Computing methodologies::Image processing and computer vision
• Online Access: https://hdl.handle.net/10356/166148
• Institution: Nanyang Technological University

Crowd counting technologies, have in recent times, seen an upsurge in popularity due to the wide range of practical applications they offer, spanning from safety monitoring to disaster management, and public space design, among others. This phenomenon has piqued the interest of a plethora of professions, including the police, civil defence force, farming communities and many more. Nonetheless, despite their impressive capabilities in handling such tasks, they often come with a set of challenges, and this has spurred the computer vision community to further invest in this field of research. Advancements in technology have led to the emergence of different types of neural network architectures, which has solved the numerous challenges faced by traditionally hand-crafted features, such as their limited learning capabilities, achieving state-of-the-art performances whilst yielding accurate results. Although these technologies are novel and revolutionary, factors like occlusion and scale variation remain a constant challenge to these crowd counting architectures. Thus, this final year project aims to study the several types of architectures used for crowd estimation, to gain a deeper understanding as to how each component within a network works. Subsequently, a network that utilises different architectures will be implemented and fused, emphasising not only efficiency but also tackling the challenges mentioned earlier, with the primary aim of providing a more accurate count of the crowd in an image. Finally, the efficacy of the implemented model will be thoroughly evaluated through experiments and evaluations, with the aim of assessing its effectiveness towards the task of crowd counting, to determine the best integration.