A Silhouette Based Technique for the Reconstruction of Human Movement

A number of promising applications have renewed researchers’ interest in the analysis of human movements. In general, motion capture could play an important role in many areas that require storing, analyzing, or reproducing the motion of human beings. Current motion capture techniques are based on intrusive sensory systems, which might be disturbing or impossible to apply in several application areas. In this paper we present a novel non-intrusive technique able to reconstruct unconstrained motion. From multiple-viewpoint images taken with an ordinary camera a 3D reconstruction is computed with a technique known as volume intersection. Motion data are acquired by fitting a model of the performer to the reconstructed volume. Data about the reconstruction accuracy achievable with our technique in a virtual environment are also provided. c © 2001 Academic Press