HMM-Based Techniques for Intent Recognition in a Simulated Realistic Naval Environment
Loading...
Authors
Gamino, Alexander J.
Issue Date
2016
Type
Thesis
Language
Keywords
Artificial Intelligence , Computer Vision , Hidden Markov Models , Machine Learning , Pattern Recognition , Robotics
Alternative Title
Abstract
Activity recognition aims to recognize the actions of one or more agents froma series of observations. Intent recognition is an area of research dedicated toautomatically detecting and predicting the intentions of agents in an environmentbefore they are realized. Intent recognition occurs in conjunction with activityrecognition, and both may use the same set of low-level features. When applyingintent recognition to simulated environments, realism is a key factor for producingsystems which are adaptable to real-life settings. This thesis explores techniquesfor creating intent recognition systems which are readily adaptable to real worldscenarios.We seek to develop methods of intent recognition for multiple ships in arealistic naval environment, where the task is complicated by realistic physics andship movements. Our intent recognition system is primarily based on HiddenMarkov Models that have been trained on large datasets of simulated scenarios,where ships are enacting the patterns which we wish to later recognize. Thesepatterns involve ships performing various maneuvers, such as blocking orramming. Based on the detection of a set of highly discriminatory featuresextracted from the data, the HMM is able to recognize patterns in order to reliablyseparate ship actions into various intent classes. Using a temporal window ofactivities, our system predicts ship intentions before they are realized, potentiallyallowing for responsive actions.
Description
Citation
Publisher
License
In Copyright(All Rights Reserved)