Smart object counter

• Main Author: Xiao, Lingyi
• Other Authors: Loke Yuan Ren
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science; Object counting; Computer vision
• Online Access: https://hdl.handle.net/10356/181135
• Institution: Nanyang Technological University

Imagine a world where object counting didn’t exist. There would be no quick way to estimate how many people are in a crowded train station, how many cars are passing through an intersection, or even how many products are left on a store shelf. We’d be stuck relying on slow, manual counts—an inefficient task for humans and impossible at large scales. In recent years, object counting methods have become more innovative, particularly to overcome the limitations of data scarcity. There are now three main types of objects counting: Few-Shot Counting, Reference-less Counting, and Text-Guided Counting. After evaluation different types of objects counting. It is observed that existing few-shot counting models struggle to generalize across diverse object classes, particularly in complex scenes with varying object sizes and densities. Therefore, this project proposes an improved few-shot counting model that incorporates multi-scale feature fusion techniques with detect and verify paradigm. Our model introduces a multi-scale feature extraction structure that improves the detection and verification processes, achieving greater adaptability to diverse object appearances, improving counting accuracy, scalability, and generalization across complex visual contexts.