Kyrollos, Daniel G., Dabkowski, Ryszard, Pejemsky, Anya, Gwynne, Steve ORCID: https://orcid.org/0000-0002-2758-3897 and Roberts, Shelley (2026) Measuring passenger touch behaviors in aircraft cabins with domain-adapted vision models. In: 2026 IEEE International Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM), Amalfi, Italy, 2026. IEEE Xplore . IEEE (Institute of Electrical and Electronics Engineers), Piscataway, New Jersey, pp. 1-6. ISBN 979-8331551759 (doi:10.1109/AI4IM69129.2026.11558203)

Preview

PDF (Author's Accepted Manuscript (IEEE)) 54338 GWYNNE_Measuring_Passenger_Touch_Behaviors_In_Aircraft_Cabins_(IEEE AAM)_2026.pdf - Accepted Version Available under License Creative Commons Attribution. Download (14MB) | Preview

Abstract

Accurate modeling of disease transmission during air travel requires empirical data on passenger behaviors, particularly fomite-mediated interactions within aircraft cabins. Traditional observation methods for quantifying fomite interactions are labor-intensive and limited in scope. We present Domain Adapted Touch (DAT) models in a two-stage training framework for detecting cabin-specific touch behaviors from high-volume video. Training data were collected via a semi-automated curation process that mined candidate events of fomite interaction and verified them manually. We first apply supervised contrastive learning on large, generic touch datasets to learn contact-sensitive representations, and then train task-specific DAT classifiers on curated cabin datasets for face, tray table, window, and seat-back interactions. Evaluated on held-out test sets, DAT classifiers out performed those trained on generic foundation representations. Average precision improved from 0.669 to 0.830 for touches to tray tables, 0.592 to 0.740 for windows, and 0.826 to 0.949 for faces, with smaller but consistent gains for seat-back touches. The results demonstrate that the two-stage framework enhances sensitivity to spatial and contextual cues in visually complex cabin environments. Functioning as a scalable “virtual sensor,” the system enables continuous measurement of rare behaviors and provides high-fidelity inputs for fomite-mediated transmission modeling. This work is part of a broader gate-to-gate system, with ongoing extensions to airport terminals, jet bridges, and aerosol-relevant behaviors to support comprehensive behavior disease risk assessment across the full passenger journey.

Item Type:	Conference Proceedings
Title of Proceedings:	2026 IEEE International Symposium on Artificial Intelligence for Instrumentation and Measurement (AI4IM), Amalfi, Italy, 2026
Uncontrolled Keywords:	modeling, contacts faces, training, printing, hands, surfaces, aircraft, aluminum, windows, air travel, disease transmission, fomite, behavioral measurement, supervised contrastive learning
Subjects:	H Social Sciences > HD Industries. Land use. Labor > HD61 Risk Management Q Science > Q Science (General) Q Science > QA Mathematics Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Faculty / School / Research Centre / Research Group:	Faculty of Education, Health & Human Sciences Faculty of Engineering & Science > School of Computing & Mathematical Sciences (CMS)
Last Modified:	04 Sep 2026 14:13
URI:	https://gala.gre.ac.uk/id/eprint/54338

Actions (login required)

View Item

Downloads