Construction of Dense 3D Ship Models from Submarine Image Collections

To create robust automated video analysis techniques for extracting, measuring, classifying, and identifying objects-of-interest, 3D object models have many benefits. Manual creation of a full range of 3D models would be a daunting task. A method for automatically generating 3D models from data gathered at sea would make populating a 3D model database practical as well as enabling updates to be generated as more information is gathered. Toyon Research Corporation proposes to implement a method for constructing a 3D model from a series of submarine-acquired 2D images. Toyon has developed industry-leading techniques for dense 3D reconstruction from moving sensor platforms. The effort will include development of methods for segmenting target ship data from marine backgrounds, automatically estimating the relationship among image viewpoints, compensating for target ship motion, and integrating results into a 3D model. A performance metric suitable for predicting automated target recognition performance will be created. The initial software implementation of the proposed algorithm will be used to assess the potential of the proposed reconstruction process and provide the basis for a recommended system design.