Abstract-Band - Fakultät für Informatik, TU Wien - Technische ...

Paul Staroch
A Weather Ontology for Predictive Control in Smart Homes
Studium: Masterstudium Software Engineering & Internet Computing
BetreuerIn: Ao.Univ.Prof. Dr. Wolfgang Kastner
In the past few years, the idea of creating smart homes has gained popularity.
A smart home possesses some kind of intelligence that allows it to support its
inhabitants. The overall goal is to increase the inhabitants' comfort while
energy use and costs are reduced as well. In the context of a project aimed at
building smart home systems, this thesis aims at constructing an OWL ontology
for weather information, containing data about both current conditions and
weather forecasts. The data described by this ontology will enable smart home
systems to make decisions based on current and future weather conditions. At
first, the thesis will determine in which particular ways weather data can be
used within smart homes. Furthermore, possible sources for weather data will
be analysed. Primary sources will be weather services that are that are accessible
via Internet. Optionally, local weather stations will be able to provide
further data about current weather conditions. A set of Internet-based sources
for weather data will be reviewed for their suitability for use within smart
homes. Afterwards, existing ontologies will be reviewed for their structure,
advantages, and disadvantages in order to acquire ideas being suitable to be
re-used. Several well-known approaches for building new ontologies from
scratch will be discussed in detail. The thesis will follow METHONTOLOGY, the
best-fitting of these approaches, to build Smart-HomeWeather, an OWL
ontology that covers both the weather data being available and the concepts
required to perform weather-related tasks within smart homes while always
keeping the possibility of simple and efficient OWL reasoning in mind. Eventually,
Weather Importer, a Java application, will be developed that gathers
data from weather services and local weather stations and transforms it to
comply with the SmartHomeWeather ontology.
Martin Stütz
Detecting Illicit Content in Picture Streams
Studium: Masterstudium Information & Knowledge Management
BetreuerIn: Ao.Univ.Prof. Dr. Wolfgang Kastner
18
Diese Arbeit beschreibt einen Algorithmus zur Erkennung von Farbbildern mit
pornographischen oder gänzlich illegalen Inhalten. Dazu kann der Algorithmus
eine Altersklassifizierung anhand von gesichtsbasierter Geometrie durchführen.
Der Algorithmus erweitert den Algorithm for Nudity Detection von Rigan Apapid
durch die Verwendung von Support Vector Machines und Methoden zur
Eliminierung von Fehlalarmen. Die Nacktheits- und Pornographieerkennung
basiert auf Hauterkennung. Mithilfe von Hauterkennung können Hautflächen in
Bildern lokalisiert werden. Größen, Formen und Distanzen der erkannten Hautflächen
zueinander werden als Features zur Beschreibung der Inhalte von
Bildern verwendet. Basierend auf diesen Features und dem Gesamtanteil von