freeviewpoint

Freeviewpoint, also known as free-viewpoint video (FVV), is a media technology that enables viewing a scene from arbitrary viewpoints after capture. It combines data from multiple cameras and, in some approaches, depth information to reconstruct a virtual representation of the scene and render novel views for interactive viewing.

Most freeviewpoint systems begin with synchronized footage from a camera array. The captured images are calibrated and often supplemented with depth maps or 3D geometry such as point clouds or meshes. This information is used in view synthesis to generate images from viewpoints between or beyond the physical cameras. Common synthesis methods include depth-image-based rendering, which shifts textures according to estimated disparity, and more general view interpolation techniques based on light fields or multi-view geometry. Some approaches model the scene as a 3D proxy and render from any virtual camera position.

Applications include free navigation in sports broadcasts, concerts, and film productions, telepresence, virtual reality, and interactive media. Real-time or near-real-time playback remains challenging due to large data volumes, computational requirements, and the need for accurate depth estimation and camera calibration. Ongoing research aims to improve handling of occlusions and artifacts, reduce holes and ghosting, and compress plenoptic data for streaming and storage.

History and development trace back to image-based rendering and multi-view stereo research from the 1990s and 2000s, with later systems explicitly marketed as freeviewpoint video in the 2010s for research prototypes and select broadcast experiments.