|My Role||Group Manager & Principal Scientist|
|Duration||September 2019 to August 2022|
|Coordinator||HFC Human-Factors-Consult GmbH, Berlin|
|Partners||Fraunhofer-Institut für Zuverlässigkeit und Mikrointegration IZM, Berlin|
|OFFIS e.V., Oldenburg|
|Die Netz-Werker AG, Berlin|
|OndoSense GmbH, Freiburg (subcontractor)|
As a co-proposer and group manager, I lead this project at OFFIS and supervise the researchers working on this project.
OmniConnect aims to track many everyday objects and build intelligent location-based services. Through learning, one such service may be to remind of a forgotten key when leaving the home. Another service may be to localize objects in the home. Residents can be tracked indirectly through their garments. One service building on resident tracking is accurate fall detection, with enough discriminative power to separate lying on the bed versus lying on the floor.
The project OmniConnect - Multiple networking of everyday objects via 60GHz label - funded by the Federal Ministry of Education and Research (BMBF) starts in September 2019. It develops a system for the simple connectivity of things by means of passive radar tags that are approx. 1cm² small and flexible. This allows such tags to be attached to many different, traditionally non-networked things, such as clothing, glasses, pens or the like. The system is intended to develop different assistance scenarios by implicit interaction with these networked things. The participating partners besides OFFIS are Human-Factors-Consult GmbH (coordinator), Fraunhofer IZM and Die Netzwerker AG.
The core of the research at OFFIS is the tracking and clustering of several tags (such as all those belonging to a person), as well as behavioral modeling and prediction. Since it is not always clear which tags belong together logically, data-driven clustering is necessary during runtime. This makes it possible, for example, to determine which garments a person is currently wearing, or to tell several people apart. In corresponding models, these correlations are learned probabilistically in order to make statements about the identity of the wearer and to derive conditions for activities. Temporal models will record and predict sequences of activities in order to remind one of a forgotten key, for example.