Video Embed
Presentation Type
Lecture

Multi-Domain Situational Awareness

Presenter
Title

Paolo Braca

Country
ITA
Affiliation
Centre for Maritime Research and Experimentation

Presentation Menu

Abstract

In order to compose an operational picture that is as complete as possible, modern surveillance systems have to integrate coherent information from a wide variety of sources. While in the past surveillance had suffered from a lack of data, current technology transformed the problem into one of an overabundance of information, leading to a vital need for automated analysis. Indeed, current surveillance sensors generate volumes of data that only a few years ago would have been inconceivable. All the processing, algorithm calibration, parameter tuning, etc., need to be as much automatic as possible. More importantly, we are in need of novel paradigms for algorithmic design. In this respect, Artificial Intelligence (AI) and Data Fusion (DF) offer an unprecedented opportunity to strengthen the technological edge; however, the risk is to elevate, at the same time, the speed of the threats we face. Indeed, the surveillance task is complicated by the diversification of threats, whose nature and origin are most often unknown. AI and DF techniques have the potential to identify patterns emerging within these very large datasets, fused from a variety of sources and generated from monitoring wide areas on a day-to-day basis, and use the learned knowledge to anticipate the possible evolution(s) of the operational picture. 

This webinar will focus on both real-world scenarios and theoretical models, spanning from the underwater to the space domain, including the analysis of scenarios with heterogeneous surveillance sensors (radar, sonar, and satellite). The opportunity will be taken to show also an application of the aforementioned techniques to a completely different scenario which is the epidemiological COVID-19 curve monitoring and forecasting. Finally, a novel theoretical analysis will be described, which provides for an approximation −asymptotically exact− of the error probability of Machine Learning (ML) binary classifiers. This last theory is specifically relevant for radar/sonar detection.