On the Analysis of Human Posture for Detecting Social Interactions with Wearable Devices

Detecting the dynamics of the social interaction represents a difficult task also with the adoption of sensing devices able to collect data with a high-temporal resolution. Under this context, this work focuses on the effect of the body posture for the purpose of detecting a face-to-face interactions between individuals. To this purpose, we describe the NESTORE sensing kit that we used to collect a significant dataset that mimics some common postures of subjects while interacting. Our experimental results distinguish clearly those postures that negatively affect the quality of the signals used for detecting an interactions, from those postures that do not have such a negative impact. We also show the performance of the SID (Social Interaction Detector) algorithm with different settings, and we present its performance in terms of accuracy during the classification of interaction and non-interaction events.