Image-Based Stress Recognition Using a Model-Based Dynamic Face Tracking System

Dimitris Metaxas, Sundara Venkataraman, and Christian Vogler

Center for Computational Biomedicine, Imaging and Modeling (CBIM), Department of Computer Science, Rutgers University

Abstract. Stress recognition from facial image sequences is a subject that has not received much attention although it is an important problem for a host of applications such as security and human-computer interaction. This class of problems and the related software are instances of Dynamic Data Driven Application Systems (DDDAS). This paper presents a method to detect stress from dynamic facial image sequences. The image sequences consist of people subjected to various psychological tests that induce high and low stress situations. We use a model-based tracking system to obtain the deformations of different parts of the face (eyebrows, lips, mouth) in a parameterized form. We train a Hidden Markov Model system using these parameters for stressed and unstressed situations and use this trained system to do recognition of high and low stress situations for an unlabelled video sequence. Hidden Markov Models (HMMs) are an effective tool to model the temporal dependence of the facial movements. The main contribution of this paper is a novel method of stress detection from image sequences of a personís face.

LNCS 3038, pp. 813-821.

Last modified: