US Patent 10257494 Reconstruction of three-dimensional video

In one embodiment, a method includes accessing multiple images representing a 360° view of a scene, each image representing a portion of the 360° scene and including a first overlap area and a second overlap area, the first overlap area corresponding to an overlap with a first adjacent image, and the second overlap area corresponding to an overlap with a second adjacent image. The method also includes, calculating, for each overlap area, a set of homographies based on one or more feature points in the overlap area. The method further includes creating a set of left images from the multiple images, the set of left images representing a first 360° view of the scene. The method also includes creating a set of right images from the multiple images, the set of right images representing a second 360° view of the scene.