Ship Shape: Three-dimensional ship modeling for submarine combat systems

The rapid identification of ships from periscope and other imagery is a capability important to a variety of submarine missions, including sea control, anti-surface warfare, battle group support, covert operations landings, and ISR (intelligence, surveillance, and reconnaissance) operations. When spotting a new vessel, the imagery taken from submarines can be useful to construct three-dimensional (3D) computer-aided design (CAD) models of ships that can be used for identification in future sightings, for computer-generated simulations, and for capabilities assessment. Currently the task of ship recognition from images requires substantial human involvement, and model building is labor-intensive. This research project seeks to demonstrate the feasibility of applying modern computer vision techniques to create a Ship Shape system that computes 3D CAD models from submarine periscope imagery of ships, with operator involvement limited to 15-30 minutes per model. We propose to leverage current computer vision techniques and demonstrate proof-of-concept algorithms for each phase of the computation, including image segmentation, sparse shape reconstruction, dense shape reconstruction, and CAD modeling with limited user interaction.