Human pose recovery and behavior analysis from RGB and depth maps
Tutor / director / evaluatorEscalera, Sergio
Document typeMaster thesis
Rights accessOpen Access
English: Robust human pose recovery and automatic behavior analysis has applications including gaming, human-computer interaction, security, telepresence, and health-care, just to mention a few. In this work, we present a generic framework for human posture analysis and gesture recognition using RGB-D representation. It encompasses the process we have undertaken to analyze human postural configurations reliability and robustness. The work ranges from the process of image acquisition, beginning in geometric models of image representation, in order to understand the power and the use of RGB-D spaces. Having described this technology, the main focus of this work is based on the location and description of human models which can represent and describe the human pose with high accuracy. We defined an accurate pose descriptor, and defined a generic framework for automatic multi-class behavior analysis. Several applications of the proposed methodology are also presented and discussed.