Building Corpora of Bodily Expressions of Affect

Studies aimed at measuring users ’ engagement with interactive products mostly rely on collecting self-report data. However, there is an increasing body of research into systems able to assess the affective state of a user from objective data such as body movements. To train such systems we need corpora of motion capture data showing people in emotionally charged situations. Here, we report on our ongoing research into building corpora of movement data. In particular, we focus on two crucial areas in the building process. Firstly, challenges related to recording corpora "in-the-wild", i.e. in ecologically valid situations. Secondly, challenges related to obtaining ground truth labels, i.e. ensuring that individual sequences of movements are labeled correctly with affective states that can be seen in them