Ein AAL-Ansatz zur Status- und Aktivitätsbeurteilung durch Do- mainexpertenwissen mit wenigen und nichtintrusiven Sensoren An AAL Approach to Status and Activity Assessment by Use of Domain Expert Knowledge based on Sparse Nonintrusive Sensors Mayer P 1 , Panek P 1,2 1 Fortec - Research group on Rehabilitation Technology, Inst. “integrated study”, Vienna Univ. of Technology 2 Ceit Raltec – Institute for Rehabilitation and Assisted Living Technologies gemn.GmbH, Schwechat, Austria Abstract. In the Austrian eHome project a prototype system for supporting independently living older per- sons has been developed. It is based on a sparse set of unobtrusive sensors and on domain expert knowl- edge from care area. A relatively small collection of core rules was found to be sufficient for a useful ro- bust and quickly installable system irrespective of the individual home setting. Introduction and Aim: In contrast to other ap- proaches the distributed e-Home system uses a sparse set of unobtrusive sensors. Core parts of the systems are: Figure 1 Main components of e-Home system. Via IP and an external alarm routing service (ARS) phone calls to carers, family members, emergency call centres can be established Methods: An architecture based on the knowledge of domain experts from care area was designed includ- ing a rule based decision taking engine (similar to ARDEN syntax). A relatively small collection of core rules is used aiming at a useful robust and quickly installable system irrespective of the individual home setting. Learnability focuses on adaptability to indi- vidual habits, initial values allow using the system right after installation. Figure 2 Layers for ADL extraction and rule based expert system Results: Evaluation of ADL (activity of daily living) recognition was done in laboratory and in a one week pilot test at an older person’s home. The preliminary findings showed good correlation between the sys- tem output and the self reported user diary. night true positive false positive false negative Self- reported Tue 16 – Wed 17 2 -- -- 2 Wed 17 – Thu 18 2 -- -- 2 Thu 18 – Fri 19 4 -- -- 1 Fri 19 – Sat 20 1.5 -- 0.5 1 Sat 20 – Sun 21 3 -- -- 3 Sun 21 – Mon 22 4 -- -- 3 total 16.5 -- 0.5 12 Table 1 ADL recognition of nightly walking to and returning from toilet. Output of rule 155 and 156 compared with notes in user diary Figure 3 Statistics for normalised activity rate in kitchen. Hourly rate shows the current activity rate, avg the average, avg-stddev and avg+stddev are the borders for deviation Discussion: The developed system worked fine. An additional evaluation of the described approach was carried out successfully in the last phase of the pro- ject in real life settings and is reported elsewhere (cf. posters P1.35 and P1.38). Acknowledgement: Research reported here was carried out in the RTD project "e-Home - Context-Aware and Distributed Embedded System for Assistive Home Technology" which is partly funded within the FIT-IT research programme (contract number 815195) by the Austrian Federal Ministry for Transport, Innovation and Technology. Project partners are Vienna University of Technology, Institute 'integrated study' CEIT RALTEC gemeinnützige GmbH, Kapsch CarrierCom AG, TREVENTUS Mechatronics GmbH. P2.2 Demographischer Wandel – Assistenzsysteme aus der Forschung in den Markt 4. Deutscher AAL-Kongress, 25.-26. Jänner 2011, Berlin