Marker-less Real Time 3D Modeling for Virtual Reality

Today, immersive environments mainly use a few 3D positions given by trackers to align the real and virtual worlds. In this paper we present an alternative approach using commodity components to achieve real time marker-less 3D modeling in virtual reality environments. The goal is to compute in real time a 3D shape of the objects and users present in the interaction space without having to equip them with markers. Having full 3D shapes opens new possibilities for full-body interactions and a tight virtual/real world integration. Data acquisition is performed through cameras surrounding the interaction space. A visual hull reconstruction algorithm is parallelized on a PC cluster to compute in real time the 3D shapes of the scene observed. Experimental results using 4 cameras and 20 processors lead to a precise 3D human model built in real time.