Particle Filtering Strategies for Visual Tracking dedicated to H/R Interaction

Summary. This paper deals with visual tracking of people from a camera mounted on a mo-bile robot in a human, cluttered, environment. Various visual cues are described, relying on color, shape or motion, together with several particle filtering strategies taking into account all or part of the measurements. These strategies enable the combination/fusion of visual cues, both into an importance function from which the particles are sampled, and into a measure-ment model serving in the definition of weights. The paper describes some prominent visual-based interaction modalities for our tour-guide robot and checks which visual cues and filtering algorithms associations best fulfill their requirements. Extensions are finally discussed.