Description
In this thesis we analyze threats from seemingly unobtrusive sensors
embedded in Smart Home devices. We explain how multiple harmless data
sources can be combined to gain personal insights into users’ lives.
Further we look into possible real world implementations of attacks on
users’ privacy. Therefore the capabilities and data availability for
different actors are considered to determine the severity of such
threats. To demonstrate a theoretical attack utilizing CO2 levels in a
room, a real-time capable machine learning model is presented.
Differentiating natural processes related to CO2 changes from human
breathing was approached by feature engineering exponential and linear
fit functions. This allows using much simpler machine learning models
than existing approaches require.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching