Collaborative Privacy-Preserving Video Analytics at the edge
Event Type
DescriptionVideo analytics plays a key role in smart cities and connected applications such as crowd counting, activity detection, event classification, traffic counting. Todays, video analytics is typically done using a cloud-centric approach where data is funnelled to a central processor with high computational power while the privacy of this data is compromised since the cloud server have access to this data. This approach introduces several key issues. Executing DNNs inference in the cloud raises privacy concerns. As an example, the use of computer vision technologies using cameras is not limited to the rapid adoption of facial recognition technology but also extends to facial expression recognition, scene recognition, etc. These developments raise privacy concerns regarding collection and use of sensitive personal data. These concerns can grow to an extent that regulators and authorities take serious actions with regards to these technologies.
As an example, recently, San Francisco banned facial recognition technology. Most of the current privacy-aware video streaming approaches involve denaturing which means content-based modification
of images or video frames, guided by a privacy policy Motivated by the aforementioned challenges, we use our build on top of our previous works to envision a new paradigm for collaborative privacy preserving cross camera video analytics at the edge of the network where the video frames are processed in the edge and their sensitive data is filtered before sending them to a third party cloud server. We believe that such a paradigm can significantly ensures better privacy.