Traffic Condition Estimation at the Smart City Edge using Deep Learning: A Ro-Pax Terminal Case Study

Abstract

The technology and innovations introduced by information systems have revolutionized how public institutions use their infrastructures and manage their assets. Even more, the need to carry out maintenance and to guarantee adequate standards of safety, efficiency, and quality of the services provided has prompted administrations and managers to integrate, in their management plans, the use of Information Computer Technologies (ICT). They are used as a tool through which it is possible to ensure the transition between the previous management and the new needs in the context of Smart mobility and, in general, Smart City. In this sense, this paper introduces a remote monitoring system built at the port terminal of Tremestieri (located in the city of Messina, Italy) that exploits deep learning to perform an Automatic Number Plate Recognition (ANPR) of vehicles transiting the area. Using a set of cameras installed at the terminal, this system allows to compute synthetic traffic indicators (e.g., arrival distribution, travel time, platooning and queuing phenomena) that can be used to infer the traffic conditions and the overall operability of the infrastructure in a non-invasive manner, highlighting daily peak hours and trends in the use of the terminal, also achieving an estimation of congestion phases.