SOCIAL INTERACTION ANALYSIS USING A MULTI-SENSOR APPROACH

Abstract

In this thesis, we have looked at the problem of social interaction analysis from a novel multi-sensor perspective. In contrast to the earlier works, which take social interaction as one type of complex human behaviors, we model social interactions using the social signals. Additionally, we explicitly take the sensor environment into consideration, including both ambient sensors and wearable sensors. In the first work, we modelled social interactions by adopting the sociology concept ``F-formation'', which analyzes the spatial properties of social interactions in a generic ambient sensor environment. Following the first work, we studied the spatial structure of social interactions in a multiple wearable sensor environment. Finally, we combined ambient sensors and wearable sensors to investigate a special yet broad class of social interactions "presentation" in this multi-sensor and multi-modal sensor environment.