Combining Shape-from-Motion Output with Partial Metadata

When the positions of the camera are not given, shape-from-motion methods can use target feature positions identified in a sequence of images to extract the three-dimensional positions of the features and the orientations of the camera relative to the target. These methods can be applied to video sequences taken from an aircraft, but do not reveal the overall size, orientation or geographic location of the target. If the camera is combined with a system that records its location, orientation and focal length, at least approximately for a few of the images, the target size, orientation and location can be estimated In certain cases, some of this information can be estimated from the image sequence itself. This paper considers the theory and practice of combining the shape extraction with the available camera information.