Learning Privacy for the Internet of Moving Visual Things

Research topics Visual Privacy | Machine Learning | Information Theory and Processing

Description

We are gradually witnessing the emergence of massively connected systems composed of things capable of sensing, analysing, processing, and communicating data about the environment and the people in it. This so-called “Internet of Things” is being proposed as the basis of a wide range of exciting applications such as smart homes, smart cities, intelligent transportation, manufacturing, and more Many of these applications will rely on networks of visual sensors to convey user data to some service provider or entity.

However, this Internet of (Moving) Visual Things is also gradually bringing to light new privacy challenges because visual data (e.g., video footage) can include relevant information for certain applications (utilities), but may also contain user sensitive information that needs to be protected. Thus, there is an emerging need to create new technologies guaranteeing that visual data is provided in a form where only the desired task (utility) can be accomplished, fundamentally blocking all the private inferences.

Our research programme pioneers new information-theory and machine learning based utility/privacy-preserving technologies applicable to video data, that can fundamentally block private information but allow simultaneously the preservation of utility information. It is expected that this programme will lead to new privacy-aware system design principles, methodologies, and a proof-of-concept to showcase the potential of the proposed new paradigm.