148 Pages

Towards less supervision for scalable recognition of daily activities [Elektronische Ressource] / presented by Maja Stikic


Gain access to the library to view online
Learn more


Towards Less Supervision forScalable Recognition of Daily ActivitiesA dissertation submitted toTECHNISCHE UNIVERSITÄT DARMSTADTFachbereich Informatikfor the degree ofDoktor-Ingenieur (Dr.-Ing.)presented byMAJA STIKICDipl.-Ing.thborn 6 of May, 1978in Sarajevo, Bosnia and HerzegovinaProf. Dr. Bernt Schiele, examinerProf. Dr. Thad Starner, co-ethDate of Submission: 29 of April, 2009thDate of Defense: 15 of June, 2009Darmstadt, 2010D17AbstractThis thesis is concerned with scalable recognition of human activities in real-world set-tings. Research towards this aim has addressed the automated detection of Activities ofDaily Living, such as personal hygiene, eating, meal preparation, or housekeeping, as aparticularly fruitful endeavor for elderly health care. The focus of this thesis lies on twochallenges within these efforts: characterization of daily activities in sensor readings, andpractical methods to label these data. We address these challenges by investigating severalresearch directions for unobtrusive activity recognition that require only a limited numberof sensors and minimal annotation overhead.We utilize a multi-sensor approach to characterize two important aspects of activities.We use wearable acceleration sensors to infer characteristic body movements and RFIDtags in combination with RFID readers to recognize object usage during execution ofactivities.



Published by
Published 01 January 2010
Reads 8
Language English
Document size 8 MB