Metadata-oriented edge node architecture

To produce, collect and handle metadata, the edge node components are organized in three layers:

• The processing layer: Perceptual components operating on the feeds of sensors, the social network manager operating on data from social networks and the linked data manager operating on linked data.
• The metadata handling layer: Formatting the metadata and forwarding them to the edge node database for storage and streaming towards other components.
• The reasoning layer: Metadata from different streams are fused together to infer the context at the intelligent fusion manager.

The layered architecture of the edge node is shown below:

Metadata feed specification

Sensor data enter the processing layer components of SMART, resulting to metadata that populate the feeds towards the edge node database. The organization of the metadata into feeds is done at the metadata handling layer. Having considered general data standards (XML and JSON), specific data standards (MPEG-7, SensorML and RDF) and multimedia ontologies, the SMART feed representation has been consolidated at two levels:

• The description of the feeds has been specified. This description is necessary for registering a feed to the system. This resulted into classifying the feeds as physical (sensor output), virtual (produced by processing data from the cloud, e.g. social networks) and mixed (metadata produced by applying some algorithm on sensor outputs, e.g. a tracker on a camera output). The representation of the feeds is in XML, based on simplifying and modifying SensorML. The feed description is detailed in Specifying the feed in the online SMART documentation.
• The feed contents (metadata) have also been specified. This is necessary for providing metadata to the rest of the SMART system. The metadata are provided as linked data in RDF. The supported RDF formats are both XML and JSON, as discussed in Populating the feed in the online SMART documentation.

Metadata handling layer

Metadata handling involves the following steps:

• Initialization of feeds and their registration with the edge node database.
• Listing the registered feeds
• Populating the feeds, and
• Retrieving metadata from the feeds

Two interfaces have been built to provide this functionality:

• The direct access interface provides direct access to the CouchDB and uses the API of the database. It offers information in JSON. As it does not have intermediate processing stages, it is the fastest interface.
• The common interface is not database-specific, effectively hiding the database from the user and thus allowing the use of any database. It is based on Java servlets, allowing both XML and JSON document formats, as well as and human-readable time codes in the queries and data.

For more information, see the the edge node section of the online SMART documentation.