Smart surveillance application fragmentation: A PrEstoCloud oriented approach

Surveillance systems that capture video and audio in enterprise facilities and public places produce massive amounts of data while operating at a 24/7 mode. There is an increasing need to process, on the fly, such huge video and audio data streams to enable a quick summary of “interesting” events that are happening during a specified time frame in a particular location. Through PrEstoCloud Project, ADITESS aims to enable a novel and adaptive architecture that builds on top of a distributed computing paradigm and is ideal for smart surveillance systems that can utilize resources at cloud, fog and edge.

To enable this new concept and benefit from PrEstoCloud project, the fragmentation of ADITESS surveillance system is required.

Surveillance Application Processing Layers:

  • Extreme Edge Layer (EEL): Raspberry PI 3 embedded systems
  • Regional Processing Units (RPU): Intel NUC computers
  • Cloud Layer: Public or private cloud resources

Considering the three layers of processing, the monolithic surveillance application has been fragmented into microservices (fragments). In particular, the Audio Analytics application consists of two main components the Feature Extractor and the Classifier. Similarly, the Video Analytics Component is now divided into several sub-components based on the processing requirements.

Application Fragments:

  • Audio Feature Extractor: The module responsible to extract features from the real-time stream. Can only be deployed on the extreme edge (Raspberry Pi)
  • Audio Classifier: Two classification algorithms are available through this module. The lightweight version can be deployed within the entire infrastructure while the heavyweight cannot be deployed on the extreme edge.
  • Video Motion Detector: The first level of real-time processing of video streams. This module can be deployed on any processing layer.
  • Video Face Detector: Responsible for the detection of faces within a video stream. This module can only be deployed on RPU and higher levels.
  • Video Face Recognition: An extension of video face detector with the capability to compare detected human faces with historic data (i.e. database with people of interest). This application can be deployed like the face detector.
  • WebUI: The graphical user interface (Command and Control Center) to support the surveillance operation. This module can be deployed on a private cloud in order to be accessible by all the regions within the infrastructure.

By providing the available application fragments, a set of deployment rules and the optimization metrics, ADITESS is expecting from PrEstoCloud the efficient deployment of services by exploiting the available infrastructure resources in a way that minimizes cloud and maintenance fees.