3-D Recovery of a Non-rigid Object from a Single Camera View Employing Multiple Coordinates Representation

Abstract

This paper proposes a novel technique for 3-D recovery of a non-rigid object, such as a human in motion, from a single camera view. To achieve the 3-D recovery, the proposed technique performs segmentation of an object under deformation into respective parts which are all regarded as rigid objects. For high accuracy segmentation, multi-stage learning and local subspace affinity are employed for the segmentation. Each part recovers its 3-D shape by applying the factorization method to it. Obviously the deformed portion containing twist or stretch motion cannot recover the 3-D shape by this procedure. The idea of the present paper is to recover such deformed portion by averaging the 3-D locations of a point on the portion described by the coordinates of respective parts. The experiments employing a synthetic non-rigid object and real human motion data show effectiveness of the proposed technique.