Publication Date:
2022-05-25
Description:
Submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy at the Massachusetts Institute of Technology and the Woods Hole Oceanographic Institution July 5, 1988
Description:
This thesis develops an approach to the construction of multidimensional stochastic models for
intelligent systems exploring an underwater environment. The important characteristics shared by such
applications are: real-time constraints: unstructured, three-dimensional terrain; high-bandwidth sensors
providing redundant, overlapping coverage; lack of prior knowledge about the environment; and
inherent inaccuracy or ambiguity in sensing and interpretation. The models are cast as a three-dimensional
spatial decomposition of stochastic, multisensor feature vectors that describe an underwater
environment. Such models serve as intermediate descriptions that decouple low-level, high-bandwidth
sensing from the higher-level, more asynchronous processes that extract information. A numerical approach to incorporating new sensor information--stochastic backprojection--is
derived from an incremental adaptation of the summation method for image reconstruction. Error and
ambiguity are accounted for by blurring a spatial projection of remote-sensor data before combining it
stochastically with the model. By exploiting the redundancy in high-bandwidth sensing, model certainty
and resolution are enhanced as more data accumulate. In the case of three-dimensional profiling, the
model converges to a "fuzzy" surface distribution from which a deterministic surface map is extracted. Computer simulations demonstrate the properties of stochastic backprojection and stochastic
models. Other simulations show that the stochastic model can be used directly for terrain-relative
navigation. The method is applied to real sonar data sets from multibeam bathymetric surveying (Sea
Beam), towed sidescan bathymetry (Sea MARC II), towed sidescan acoustic imagery (Sea MARC I &
II), and high-resolution scanning sonar aboard a remotely operated vehicle. A multisensor application
combines Sea Beam bathvmetry and Sea MARC I intensity models. Targeted real-time applications
include shipboard mapping and survey, a piloting aid for remotely operated vehicles and manned
submersibles, and world modeling for autonomous vehicles.
Description:
Principal funding for this research was provided by the Sea Grant Program of the Massachusetts
Institute of Technology. My course work and early research were supported by a graduate fellowship from the Office of
Naval Research. Other significant help has come from the Monitor Marine
Sanctuary Program of the National Oceanic and Atmospheric Administration.
Keywords:
Stochastic analysis
;
Scanning systems
Repository Name:
Woods Hole Open Access Server
Type:
Thesis
Format:
application/pdf
Permalink