As content generation and diffusion in video domain increase, the need for efficient tools to analyze, identify, search, and retrieve this content becomes more obvious. The problematic of this thesis is directly related to these needs, with the consideration of two particular applications: people counting and facial expression recognition.
Based on previous work related to face detection in video images realized in our laboratory, we go further in this thesis by developing a robust and efficient face tracking relying on the combination of kernel tracking and kalman filter, which is the core of our people counting system.
Moreover we aim at performing a precise segmentation of face in order to recognize the face expression. To this aim, we envisage to develop a novel approach which consists in achieving facial expression analysis on the basis of both permanent facial feature analysis (brows, eyes, mouth) and transient facial feature analysis using facial feature points.