Finding context attributes

Using spatial indices to form high-level context attributes:

Hierarchical attributes:

Nowhere (or everywhere) -> Garching -> FMI -> 1st floor -> AR lab

The SR graph nodes have a list of these attributes (if list is empty -> Nowhere).

These attributes group different SR subgraphs together into "domains".

ToDo: Are domains and supernodes the same?

Model bootstrapping mechanism

The context is used to specify a high-level spatial information combined with the knowledge of the current infrastructure. This information is used to optimize the search for new relations in the SR graph because only the current neighborhood defined by the attributes must be considered.

The initial context is "nowhere". The Ubitrack system broadcasts a "HelpMe" request to all nodes in the current subnet (technical details about building subnets not important here). This is realized by "context estimation components" (services). They are capable of broadcasting the messages as well as setting the attributes respectively.

"HelpMe" request

It contains: - My local ID (unique in domains, must it be unique globally ?) - My current context, modeled by attributes - A list of my configurable context estimation services

Possible answers

If the request contains no context information, an "initial context estimation" service is contacted. It returns following information:

- If context can be derived (e.g. from the fact that the connection has been established), return it with a list of services which can be contacted to refine the context. - If not, return configuration data for the local context estimation service. If no configuration is available, return the old context with empty list.

Refine context

If new context is returned with a list: - Contact the new services from the list until the context cannot be refined anymore.

If configuration is returned: - Reconfigure local services. Estimate the context yourself. If not successful, use old context.

If old context is returned: Use it as current context.

Initial context service

This service must be found iff no context is available.

Model anonymous objects

The new concept of the context is used when new objects must be identified in the SR graphs.

Picture here !

X is an object tracked by A Y is an object tracked by B

The "real" ID of X is M2, the one of Y is M2. The question is how to find the correct IDs and assign them to the anonymous objects.

Ubitrack must consider all measurements to find possible correlations between X and M2 as well as between X and M1. This is inefficient in great graphs. So we consider the spatial high-level context of both subsystems to get only the relevant measurments.

Find correlations

To fine the correlations we must probably consider following properties of the objects:

velovity, acceleration turnarounds

First derivation to eliminate static edges.

ToDo: is this working? -> see literature ToDo: more properties to consider?

Example

User setup: GPS tracker, WLAN

User exits at bus station. Send "HelpMe" request with no contect. Return: configure your GPS tracker if you have one and send information to service blablubb.

Blablubb returns context "You are in Garching".

ToDo: to be refined

-- FranzStrasser - 19 Feb 2004