What Is the Visual Hull?

In computer imaging, a visual hull is the three-dimensional (3D) shape of an object that is extrapolated from multiple two-dimensional (2D) images taken at different angles around the object being approximated. Surface data about the shape of an object is obtained by tracing the contour of an object in an image, essentially creating a silhouette of the object with no specific inside texture or detail. A collection of silhouettes, all extracted from images taken at different angles, are assembled together in 3D space and the area between known contour points is interpolated to form a 3D object that has the general 3D outline of the actual object, though perhaps without as much specific detail. The process used to create a visual hull, also known as shape-from-silhouette (SFS), can be faster, less processor-intensive and less expensive to implement than some stereoscopic techniques for capturing 3D motion or detecting the shape of 3D objects. Some of the applications that use the visual hull include computer vision obstacle detection, motion capture for medical or other analytical purposes, and virtual 3D object scanning when SFS is performed in highly controlled conditions.

The process of forming the visual hull of an object from a set of images involves isolating the silhouette of the object from the background in the images. The exact location and orientation of the cameras used to acquire the images also are important to the process. In each image, a straight path is made from the viewing plane of the image to the space of the scene and ending on the contours of the object being imaged. This is done for each image and the area where each of the paths, which resemble cones in a 3D environment, cross gives a very rough, block-like volume that contains the object within the dimensions of the scene. For some applications, such as computer vision, this information is enough to allow for basic obstacle avoidance.

The silhouettes can be further refined so smaller geometric details are translated to the visual hull. These can include holes in the object, as might occur if the visual hull was being constructed from images of a human standing with legs apart or arms outstretched. One attribute of an object shape that cannot be accurately captured with SFS techniques is a concave surface, because it does not contribute to the silhouette.

The SFS technique for creating the visual hull of an object can be incredibly detailed and accurate if refined algorithms are used in conjunction with controlled conditions to create the source images. These conditions can include a single, consistent light source, a static and measurable background, and cameras that are exactingly calibrated. Given these conditions, very precise 3D models of objects can be constructed, and motion capture can be performed without the need for markers, tracers or special equipment beyond cameras.