On 3-D scene interpretation from F-S sonar imagery

Sonar imaging systems, including the recent generation of high-frequency 2-D forward-scan video cameras, are the most effective modality for underwater operations in highly turbid conditions. Automatic processing of sonar imagery to recover information about the 3-D environment enhances utility and performance and enables new capabilities in the operation of autonomous undersea vehicles. This paper is about the reconstruction of a 3-D scene from various visual cues in 2-D forward-scan sonar imagery. We show how the sizes, contour shapes and relative positions of various 3-D objects in the scene can be determined from both radiometric information in a single image to detect object and shadow regions, and various other geometric cues. Measurements from certain common auxiliary sensors, e.g., inclinometers and (or) altimeters, can be utilized to improve accuracy. Performance is assessed based on ground-truth in experiments with both synthetic data, and real images through comparison with estimates from an independent technique.