Acquisition of 3-D surface shape of human body from monocular images without pose estimation

In this paper we present a novel method to acquire the 3D surface shape of a human body from video/images captured by a single camera. Unlike existing methods where view or motion restriction is imposed or posture estimation or fitting is required, the proposed method estimates the human surface shape by computing the shape parameters directly from features of body parts in the images, assuming body proportion and camera focal length are available (which also can be estimated from monocular images). In this method, each body part is considered as a rigid 3D object derived from a cylinder and is dealt with independently to avoid the deformation complexity of a soft object and the pose uncertainties of an articulated object. Existing algorithms for recovering the shape of a cylinder from monocular image(s) are not directly applicable for such applications as they require the cross-section contours of a cylinder. A method is proposed in this paper that can recover the shape of a cylinder base on just 2D bounding boxes. This method is further extended for recovery of the human body shape. The proposed system is applied to both synthetic and real video/images and has produced promising results where the acquired body shape resembles the input human figure.