Parameter Study on the Use of Artificial Intelligence to Optimize Response to Unattended Bags to Increase Airport Security

Abstract

As a vulnerable transportation hub, the airport can become the target of an attack at any time. Airport security, therefore, touches a fundamental aspect of our society: moving without fear. The attackers' methods become increasingly sophisticated. It is therefore essential to react quickly and adequately through unforeseen events. Every piece of unattended baggage --- a commonly occurring incident --- potentially poses a threat. The digitalization of airports has been advanced over the last decade. This circumstance opens the possibility to a broader use of Artificial Intelligence. AI is already being successfully used in individual areas of an airport. These range from intelligent video surveillance through border control to monitoring high-security areas. In this work, an exemplary environment was examined depicting the actions necessary to neutralize an unattended piece of baggage (\eg a suitcase). The AI method Reinforcement Learning, especially the Deep Q-Network method, was used to train an agent to solve the challenge of choosing an optimal sequence of actions. This special method gives rise to a set of parameters, namely learning rate, batch size and the number of iterations. By means of a parameter study, a set of parameters was searched for enabling the agent to adequately solve the challenge.