Detection and Classification of Human Activities using Distributed Sensing of Environmental Vibrations

The recognition of human activities in a smart home is an essential prerequisite in order to derive typical behaviors and needs of the inhabitants and adapt the functions of the smart home to them. Different sensor modalities, such as video or audio data in combination with machine learning methods can be used for this purpose. However, the use of video and audio data is associated with a strong infringement on the privacy of the inhabitants. In this paper, we present an alternative approach that uses vibrational data that is acquired by stationary wall-mounted sensors to detect a specific set of inhabitant activities using machine learning. We compare different neural-network based time series classifiers and show that is possible to detect the selected activities with up to 95% accuracy.