Decentralized Detection of Group Formations from Wearable Acceleration Sensors

We propose an approach for detecting collective behavior patterns using body-worn sensors in a decentralized, online manner. To reduce complexity, we introduce a set of collective behavior primitives as building blocks, adequate for detection by sensing and communications means. With this, we present a generic distributed signal processing procedure suitable for a decentralized detection of these behavior blocks. In an experiment, we verify relevant elements of this procedure by evaluating towards which extend body-worn acceleration sensor scan be used to infer whether a set of people is walking together as a coherent group. Early results discussed demonstrate the feasibility of our approach, leading to an outline of next steps for reliable detection of collective behavior patterns in real-life.