Attire insensitive person re-identification

Abstract

Person re-identification (re-ID) has emerged as a critical discipline within the realm of computer vision. The goal of person re-ID is to accurately identify and track individuals across multiple video feeds from surveillance cameras. Many traditional person re-ID models focus primarily on capturing attire features. As a result, most attire-focused re-ID methods have difficulty tracking individuals when they change attire or blend into a crowd of people wearing similar attire. To overcome these issues, more research on person re-ID is focusing on shifting the emphasis away from clothing to more permanent and distinct physical features such as body shape, skin colour and other soft biometrics. This report explores different attire-insensitive person re-ID approaches and addresses the challenges associated with the traditional approaches, particularly in the clothes-changing scenario. Through the consolidation and evaluation of various attire-insensitive methodologies, the project seeks to bridge the existing research gap and improve the accuracy of re-ID algorithms. By investigating techniques to perform re-ID on individuals who have changed clothing, this research contributes to advancing the reliability and effectiveness of person re-identification methods, thereby addressing key concerns in security and public safety.