Applying artificial intelligence to ground and flight operations is a combined challenge and opportunity to aviation. At Zurich Airport this year, Swiss International Air Lines conducted a three-month trial of AI to count passengers when boarding the aircraft. We asked Pete Steinmann, director of operations development and strategy at Swiss, about the trial, how it went, what the airline learned, and what’s next for AI.
Installing a camera temporarily within selected aircraft cabins, Steinmann tells us, “Swiss tested the digital passenger count for over three months as part of a proof of concept. We recorded boarding sequences on all aircraft types under various conditions and used them to develop an initial software prototype that uses artificial intelligence to count our guests as they boarded.”
Ensuring an accurate count is, of course, operationally critical. As standard practice, automatic or manual boarding gate counts are cross-referenced with crew counts. Every so often, however, a high-profile example of passengers slipping through gaps occurs, which is embarrassing — and costly — for airlines.
“The purpose of digitising this process is to increase safety and at the same time relieve the cabin crew,” Steinmann explains. “The recordings are used to train the AI for the boarding process. For example, it must be able to recognise when a person is carrying a child in his or her arms and also function properly in difficult lighting conditions. The findings are constantly analysed in order to continuously increase reliability. As this was a test phase, however, the cabin crew carried out a regular, manual passenger count in parallel as usual during boarding.”
With privacy and the safety of personal data very much a concern for many passengers, and regulators keen to ensure data security, Swiss was careful to carry out its trials in full compliance with the European Union and Swiss data protection regulations (GDPR and nFADP, respectively, which are largely harmonised).
Passengers were, of course, informed of the trial prior to embarkation, and Steinmann notes “no special reactions” from travellers.
The recordings were used only for passenger count purposes, were limited to the aircraft door area, were video-only, and systems were designed not to identify persons when analysing the video — which was then deleted.
“The test operation was successful: we achieved the targets for counting people at the aircraft door, and in principle the software reliably counts the number of boarding passengers. Currently, we are still working on the division of people into groups: guests, crew and airport staff,” Steinmann says. “What is important to understand is that in each repetition of the AI learning we are progressing towards our goal, which at the same time means that it will take some time until all desired parameters are fully achieved.”
“During the proof-of-concept we have committed to a certain computer vision model. Whether we stay with this model or switch to another model in the future depends on the experience gained during the development of the solution system. Currently, it would be too early to specify and therefore we do not want to communicate any technical details before we have finally determined our model based on the learning successes of the AI.”
With Swiss currently tendering for its IT requirements, the airline is keeping its implementation cards close to its chest.
Overall, though, this proof-of-concept was indeed a success. “As a result of the successful test phase, SWISS intends to replace the manual counting process at boarding with this automated and digital passenger count with AI,” Steinmann concludes.
Author: John Walton
Published: 14 November 20203
Feature image courtesy of Swiss International Air Lines
;
