Group Behavior Recognition using Dynamic Bayesian Networks

In this PhD thesis we analyze the concepts involved§in the decision making of groups of agents and apply§these concepts in creating a framework for performing§group behavior recognition. We present an overview of§the intention theory, as studied by some great§theorists such as Searle, Bratmann and Cohen, and§show the link with more recent researches. We study§the advantages and drawbacks of some techniques in§the domain and create a new model for representing§and detecting group behaviors, the aim being to§create a unified approach of the problem. Most of§this thesis is consecrated in the detailed§presentation of the model as well as the algorithm§responsible for behavior recognition. Our model is§tested on two different applications involving human§gesture analysis and multimodal fusion of audio and§video data. By means of these applications, we§advance the argument that multivariate sets of§correlated data can be efficiently analyzed under a§unified framework of behavior recognition. We show§that the correlation between different sets of data§can be modeled as cooperation inside a team and that§behavior recognition is a modern approach of§classification and pattern recognition.