Multi-aspect Underwater Scene Understanding

Abstract

Using high-resolution side-looking sensors such as SAS, it is possible to map large areas of sea-bedwith finely detailed imagery. If given the capability of accurately characterizing bottom sediments and seabedcomposition using these high-resolution sensors, then environmentally-dependent applications such asthe prediction of mine-hunting effectiveness in shallow water and sea lanes, environmentally-adaptive automatedtarget recognition (ATR), and autonomous in situ mission planning could be realized in an effectivemanner. The overall goal of this proposed research is to develop an autonomous underwater scene understandingsystem that analyzes collections of multi-aspect side-look synthetic aperture sonar (SAS) imagerygenerated during a sonar survey. In this proposal, scene understanding refers to identification of the sea-floortype and estimation of the relevant descriptive parameters for that sea-floor type. This goal will be achievedthrough investigation of the following three research objectives:1. Develop physics-based parametric sensing models and associated hierarchical Bayesian parameterestimation frameworks in order to characterize multiple seabed types with invariant parameters. The parametersof interest are invariant to sensing geometry including factors such as aspect angle, height of the SASarray off of the sea-floor, range, etc. The sensing models would be valid over typical sonar ranges and for avariety of sensing aspects.2. Investigate Bayesian approaches for combining information obtained from multiple passes, ranges,and aspects to describe spatial regions and the associated sea-bed types.3. Leverage previously-developed algorithms for feature extraction and sea-bed segmentation to developan autonomous end-to-end system for underwater scene understanding.We anticipate that the proposed work will advance methods for underwater scene understanding andbathymetry estimation and can be used in conjunction with specific goal-oriented methods to enhance performance.For example, within a target detection system, the proposed work can be used to estimate minedetectionperformance in the given environment or can be used to invoke context- or environment-specifictarget detection algorithms.

Document Details

Document Type
DoD Grant Award
Publication Date
Mar 03, 2017
Source ID
N000141712271

Entities

People

  • Alina Glenn

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Florida

Tags

Readers

  • Coastal Oceanography
  • Computer Vision.
  • Neurological Diseases/Conditions/Disorders

Technology Areas

  • AI & ML