Visual hull

Content sourced from Wikipedia, licensed under CC BY-SA 3.0.

Visual hull is a simple way to create a rough 3D shape of an object from silhouettes taken from several viewpoints. It was introduced by A. Laurentini. The method assumes you can separate the object from the background in each image, producing a silhouette—the 2D outline of the object.

Each silhouette defines a cone that starts at the camera and passes through the silhouette. If you intersect all these cones from the different views, you get the visual hull—a bounding volume that contains the real object. This hull can be useful for rendering from new viewpoints, because graphics hardware can work with the volume without needing a perfect surface model.

In some devices, corner cameras with infrared LEDs can capture silhouettes or shadows to help reconstruct a convex hull. The same idea is used in telepresence experiments: multiple cameras capture a person and their interactions to build a 3D volume in real time, enabling 3D collaboration while reducing computation. This approach can even let consumer devices serve as tools for 3D interaction.