Surface Reconstruction from Point Cloud of Human Body by Clustering

This paper proposes a method in which the body surface is reconstructed from a point cloud (set of points) on the human body acquired by a device such as a laser range finder. The method constructs the surface appropri-ately for the human body by clustering the point cloud at body sections such as the chest and the upper arm. The clustering is performed as follows. A simple human model composed of cylinders, ellipsoidal columns, and hemi-spheres is assumed. The posture is estimated by matching the human body model to the point cloud data. Based on the estimated posture, the point cloud is assigned to body sections. Since the posture is inferred from the point cloud, the method is effective not only for an upright standing posture, but for point cloud data obtained from an arbitrary posture. Experiments were performed to apply the method to simulation data and measured data, and it was shown that a realistic surface could be reconstructed. The surface can be reconstructed for postures other than the upright stand-ing posture. The effectiveness of the proposed method i