Probabilistic graphical models for human interaction analysis

The objective of this thesis is to develop probabilistic graphical models for analyzing human interaction in meetings based on multimodel cues. We use meeting as a study case of human interactions since research shows that high complexity information is mostly exchanged through face-to-face interactions. Modeling human interaction provides several challenging research issues for the machine learning community. In meetings, the feature sequence extracted from each participant is multimodal. Modeling human interaction involves simultaneous recording and analysis of multiple multimodal streams. These streams may be asynchronous, have different frame rates, exhibit different stationarity properties, and carry complementary (or correlated) information. In this thesis, we developed three probabilistic graphical models for human interaction analysis. The proposed models use the "probabilistic graphical model" formalism, a formalism that exploits the conjoined capabilities of graph theory and probability theory to build complex models out of simpler pieces. We first introduce the multi-layer framework, in which the first layer models typical individual activity from low-level audio-visual features, and the second layer models the interactions. The two layers are linked by a set of posterior probability-based features. Next, we describe the team-player influence model, which learns the influence of interacting Markov chains within a team. The team-player influence model has a two-level structure: individual-level and group-level. Individual level models actions of each player, and the group-level models actions of the team as a whole. The influence of each player on the team is jointly learned with the rest of the model parameters in a principled manner using the Expectation-Maximization (EM) algorithm. Finally, we describe the semi-supervised adapted HMMs for unusual event detection. Unusual events are characterized by a number of features (rarity, unexpectedness, and relevance) that limit the application of traditional supervised model-based approaches. We propose a semi-supervised adapted Hidden Markov Model (HMM) framework, in which usual event models are first learned from a large amount of (commonly available) training data, while unusual event models are learned by Bayesian adaptation in an unsupervised manner.