DETECTION ET CATEGORISATION D'OBJETS EN MOUVEMENT DANS UNE VIDEO
Youssef Zinbi · HAL (Le Centre pour la Communication Scientifique Directe) · 2009
In the context of video analysis, it is important to have intelligent and fast segmentation methods to provide a quick overview of the content of movies. As part of this thesis, we are particularly interested in the problems of extraction and categorization of video objects. For extraction, we propose to use the global approach based asset outlines areas that can quickly locate objects of interest. For this, we used segmentation criteria that take into account the homogeneity and the perceptual attributes to define a competition between the region of interest and background. To improve the method of detection and object tracking, we extended the energy formulation of our model of global active contours including an additional force from the optical flow computation. In the second part, we address the problem of analysis of human behavior (movement and gestures) in video sequences. The goals are multiple. The term "analysis" here concerns the extraction of low-level, such as the silhouette of the person, the location of his face and the extraction of different facial components information. On the other hand, we propose a method of categorization that facilitate spectral data reduction and dimensionality of the data and the interpretation of gestures and human behavior. This is the classification of facial expressions and recognition tasks (walking, running etc.), postures (standing, squatting, etc.) people.