Smart cities are nowadays producing a massive amount of data from environmental cameras, microphones and other sensors that remain unprocessed. Therefore, many AI-based solutions are recently emerging to support new smart cities’ applications, from dangerous event detection to statistics extraction for decision making. However, these data may contain sensitive personal information, calling for responsible AI solutions. We proposes some privacy-preserving AI-components enabling responsible use of this data, by preventing unauthorised usage of personal data at both data storage and transmission under the context of Smart Cities. The proposed technologies are developed for the EU H2020 MARVEL project. Both video and audio anonymisation components are deployed at the edge level, enabled by a model compression component for complexity reduction. We discuss each component’s technical challenges, current progress, and future directions.
Our solutions answer the need to process big data from smart city sensors while at the same time guaranteeing citizens' privacy preservation.
Elementi di innovazione
The main innovation consists in the ability to run complex AI models at the edge in resource constrained devices (limited in terms of memory, processing and communication capabilities and battery powered). At the same time we optimized and tested advanced state of the art techniques for audio and video anonymization on smart city multimodal data.
Impatti / Risultati attesi
The main impact is on the ability of Public Administration to make decisions based on data analytics while guaranteeing privacy preservation for the citizens. Furthermore, law enforcement agency can benefit from AI applied to real-time data to be adviced of dangerous events requiring their fast intervention.
The prototype libraries are under test. It is a research product, not a commercial one.
Elementi di replicabilità
All results are published and described as to be repeated. Software developed is uploaded in public repositories.