Collecting Datasets from Ambient Intelligence Environments

This paper describes a methodology and lessons learned from collecting datasets in Ambient Intelligence Environments. The authors present considerations on how to setup an experiment and discuss decisions taken at different planning steps, ranging from the selection of human activities over sensor choices to issues of the recording software. The experiment design and execution is illustrated through a dataset involving 150 recording sessions with 28 sensors worn on the subject body and embedded into tools and the environment. The paper also describes a number of unforeseen problems that affected the experiment and useful considerations that help other researchers recording their own ambient intelligence datasets.