Head pose estimation and attentive behavior detection

Abstract

In this thesis, we propose two methods to detect a frequent change in focus of human attention (FCFA) from video data. In the first method, called the head pose estimation (HPE) method, we propose an algorithm developed from ISOMAP to learn a unified embedding space for different head poses. A non-linear person-independent mapping system is then proposed to map new frames or sequences into this space where head poses can be easily obtained. An entropy-based classifier is then proposed to detect FCFA behavior. In a second method, called the cyclic pattern analysis (CPA) method, we propose to use features extracted by analyzing a similarity matrix of head pose. Further, we present a fast algorithm which uses the principal components subspace to measure the self-similarity. A frequency analysis scheme is proposed to find the dynamic characteristics of similarity matrices. A nonparametric classifier is used to classify these two kinds of behaviors (FCFA and focused attention).