Hybrid body representation for integrated pose recognition, localization and segmentation

Abstract

We propose a hybrid body representation that represents each typical pose by both template-like view information and part-based structural information. Specifically, each body part as well as the whole body are represented by an off-line learned shape model where both region-based and edge-based priors are combined in a coupled shape representation. Part-based spatial priors are represented by a “star” graphical model. This hybrid body representation can synergistically integrate pose recognition, localization and segmentation into one computational flow. Moreover, as an important step for feature extraction and model inference, segmentation is involved in the low-level, mid-level and high-level vision stages, where top-down prior knowledge and bottom-up data processing is well integrated via the proposed hybrid body representation.