AI and Cloud for a Smart City that improves road safety

Road safety is a topical issue in Italy, prevention and control are increasingly fundamental to reduce accidents on our roads which unfortunately represent one of the main causes of death among young people, to the point that the Ministry of Infrastructure and Sustainable Mobility has set up an ambitious national plan to reduce fatalities and injuries by 50% by 2023.

It is in this context that the Municipality of Monserrato, a town of about 20,000 inhabitants belonging to the Metropolitan City of Cagliari, has invested in the implementation of the innovative Safespotter project to develop a platform in order to improve the safety of its roads and affect the number of road accidents.

The goal of the project was the creation of an integrated system based on Artificial Intelligence, aimed at reducing the intervention times of the traffic police in case of accidents through the acquisition of video images collected along some key points of the municipal road system and their real-time analysis, in order to identify potentially dangerous situations and recognize accidents.

Lutech, together with Arkys s.r.l. and the Department of Mathematics and Computer Science of the University of Cagliari, was awarded the contract for the construction of the project, launched by the municipal administration of Monserrato and financed by Sardegna Ricerche to develop a platform to improve road safety and affect the number of road accidents.

The result of this collaboration is a platform made up of an integrated system of connected intelligent street lamps, each equipped with an independent AI unit, capable of both detecting traffic anomalies and harmful behaviors, and identifying accidents, recognizing their severity. Furthermore, the system implements visual and acoustic signaling and warning mechanisms through luminous signs and acoustic diffusers, with the aim of informing approaching vehicles of the need to slow down or stop immediately, reducing the consequences of the main event and avoiding the occurrence of secondary accidents.

The system just described, called "first level", was accompanied by a "second level" system in the AWS cloud with two main tasks:

1. exploit in real time the information coming from the network of street lamps to carry out a territorial control of traffic (taking into account elements such as: traffic status, dangerous behavior detected, meteorological conditions, local signals, historical data, etc.), in order to predict the situations of imminent risk, as well as to globally manage the individual alerts received, propagating them where necessary, requesting assistance and promptly communicating them to motorists via messaging services;
2. benefit from the characteristics of elasticity and scalability of cloud services for offline training of models: this device allows you to reduce processing costs thanks to the cost-per-consumption model, providing, if necessary, virtually unlimited computational resources for the time required from processing.

Among the main AWS cloud services that have been used:

• Amazon SageMaker, which allows you to quickly and easily train and deploy machine learning models at scale
• Amazon Rekognition, for the analysis of video streams
• AWS Deep Learning, for deep learning with the use of the most popular frameworks