CS MSc Thesis Presentation 12 February 2026

Thursday, 12 February there will be a master thesis presentation in Computer Science at Lund University, Faculty of Engineering.

The presentation will take place in E:2116.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname [dot] lastname [at] cs [dot] lth [dot] se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions for opponents are found here on the LTH thesis project page.

15:15-16:00 in E:2116

• Presenter: Elin Persson
• Title: Evaluation of Transformer Architectures for Small Thermal Image Classification in Security Applications
• Examiner: Eren Aksoy
• Supervisor: Pierre Nugues (LTH)

As security systems become increasingly sophisticated, accurate human detection is critical to prevent costly false alarms. Thermal cameras, which produce low-resolution grayscale images through white-hot mapping, make it challenging to distinguish humans from large animals. Axis Communications currently faces this issue, as their existing classifier frequently misidentifies humans, triggering unnecessary alerts. This thesis investigates transformer-based architectures as an alternative to improve classification accuracy on small thermal images. We compare several transformer models, benchmark them against convolutional neural networks (CNNs), and explore how different architecture designs affect performance. Our study highlights the trade-off between model size and accuracy, with a focus on solutions that are lightweight enough for embedded systems. The results demonstrate a model that reduces false alarms while maintaining computational efficiency, offering a practical improvement for real-world security systems. This work contributes to the understanding of human detection in thermal imagery and provides guidance for designing efficient, high-accuracy classifiers in constrained environments.