Person re-identification of people and their luggage

Student thesis: Master thesis (including HD thesis)

  • Simon Gørtz Flou Nielsen
As travel becomes more and more common, the use of Re-identification systems become more common. For surveillance or queue estimation, Re-identification is a versatile and useful tool. By using Re-identification methods e.g. airports can keep track of passengers and their queue times for boarding and unboarding planes. For a network to distinguish a passengers from each other, a system specifically trained and developed for that purpose needs to be implemented. That is the purpose of this paper, to research person Re-identification within this context. This paper explores a dataset structure containing two bounding boxes, one for the person and another for the luggage they are carrying.
This dataset was used throughout the paper, for testing multiple ways of utilising it for Re-identification. One such method tested was a Dual-TransReID network, which reported an accuracy of \textit{95\%}. This is an improvement to the baseline test of a single TransReID network, which scored an accuracy of \textit{90\%} with a single bounding box containing both person and luggage.

Code base can be found at: https://github.com/MrFlou/Multi-Net-ReID
LanguageEnglish
Publication date2 Jun 2022
Number of pages60
ID: 471903180