Algorithms for Stereo Image Creation from a Sequence of Two Dimensional Images

We propose to develop an algorithm capable of autonomously converting standard 2D space surveillance imagery sequences into 3D stereo movies. We will extend and integrate our existing algorithms, which have already been demonstrated to handle space surveillance data, to build a system that requires no sensor or hardware upgrades. The key to automating the solution is to determine the pose of the imaged object from one frame to the next. We will use existing feature-based shape-from-motion code to solve for this unknown, and then use that information to select stereo image pairs whose displacements are compatible with binocular human vision. Because of the low frame rates of existing space surveillance sensors, we realize that many data collections may not contain suitable stereo pairs. To handle such situations we propose a novel technique to generate intermediate frames based on 3D information extracted from the image sequence. The goal is to produce a high-quality 3D stereo movie at video frame rate, even from even relatively sparse image data. For Phase I we will deliver a working prototype system stable enough for Air Force use on research and special analyses missions.