Title :
A POMDP for multi-view target classification with an autonomous underwater vehicle
Author :
Myers, Vincent ; Williams, David P.
Author_Institution :
Defence R&D Canada, Halifax, NS, Canada
Abstract :
A partially observable Markov decision process (POMDP) is proposed to perform multi-view classification of underwater objects. The model allows one to adaptively determine which additional views of an object would be most beneficial for reducing classification uncertainty. Acquiring additional views is made possible by employing a sonar-equipped autonomous underwater vehicle (AUV) for data collection. The POMDP model is validated using real synthetic aperture sonar (SAS) data collected at sea, with promising results. The approach provides an elegant way to fully exploit multi-view information in a methodical manner.
Keywords :
Markov processes; image classification; object recognition; remotely operated vehicles; sonar imaging; synthetic aperture sonar; underwater vehicles; autonomous underwater vehicle; classification uncertainty; data collection; multiview target classification; partially observable Markov decision process; synthetic aperture sonar; underwater objects; Adaptation model; Markov processes; Robot sensing systems; Shape; Synthetic aperture sonar; Automatic Target Recognition; POMDP; Synthetic Aperture Sonar;
Conference_Titel :
OCEANS 2010
Conference_Location :
Seattle, WA
Print_ISBN :
978-1-4244-4332-1
DOI :
10.1109/OCEANS.2010.5664609
