3D Human Limb Detection using Space Carving and Multi-View Eigen Models

In this paper, we integrate space carving and eigen detection methods to develop a bottom-up 3D human limb detector. We model the body in terms of its constituent body parts; here we focus on the head, lower arms, upper arms and calves. For each body part, we build a multi-view eigen model that combines image views from multiple calibrated cameras. This approach is much more constraining than the conventional multiple single-view eigen models and provides coarse 3D pose information. We use ideas from space carving using multiple silhouette images to constrain the volume of our search for the body part locations. We have applied the method to detect the body parts of a subject in long test sequences. The approach provides bottom-up in-formation that supports the automatic initialization of a full 3D human body model.