SGEER: Exploratory Research on Analyzing Forward-Scan Sonar Video Imagery by Computer Vision Techniques

University Of Miami, Coral Gables FL

Investigators

Abstract

This one-year SGER project will explore mathematical models for the application of structure from motion methods to acoustic video imagery. Computer sonar processing has been dominantly limited to classical single frame methods for scene/target detection, segmentation, classification and identification by exploiting 2-D texture or object/shadow shape cues. Little, if any, work has explored the potential use of multiple views for 3-D target reconstruction, which is becoming rather important as high-frequency acoustic cameras producing high-resolution video in low visibility conditions are becoming increasingly more available. The specific goals are to explore sonar projection and motion models in the formulation of various 3-D shape and motion estimation methods, derivation of model-based solutions, identification of some promising approaches for devising computer video analysis and interpretation techniques, and more generally developing a better understanding of the complexities in the processing of sonar data. The research project's broader impacts comprise establishing the foundation to realize critical capabilities in automated underwater mapping, image-based positioning and navigation of unmanned underwater vehicles, vehicle guidance and control, target classification and recognition, etc., eventually enabling more effective and efficient day-today deployment of underwater imaging platforms for a range of diverse applications, including benthic scientific research and exploration. Furthermore, technical developments will lead to 1) enhancing the sonar video technology for optimum performance as predicted by analytical results, and 2) promoting collaboration between vision and sonar communities with more diversified approaches that are necessary to establish sonar video processing as a more mature discipline.