The system goal is to create a distributed platform, based on neural networks trained through federated learning, collecting data from a fleet of "instrumented" vehicles with sensos able of acquiring information on the state of health of the road infrastructure, Once collected data are stored and centralized for elaboration thus reducing maintenance costs of the infrastructure.
Hybrid convolutional and recurrent neural networks, equipped with attention layers, have proven to be extremely useful in carrying out automatic recognition of object instances in a scene monitored using different types of cameras (optical, thermal and multi-spectral) or by LIDAR-based sensors allowing the presence of objects and semantically complex situations to be detected and identified, automatically and without human support.