Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance
Adaptive visual object tracking (VOT) is one of the fundamental tasks in machine vision, with active research and far-reaching implications. Bayesian methods are commonly used in adaptive VOT. However, we propose that the current tendency is to restrict the inference to a subtask (e.g. classificatio...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Language: | English |
| Published: |
Department of Electrical Engineering
2018
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867614088197046272 |
|---|---|
| access_status_str | Open Access |
| author | Bradshaw, Charles |
| author2 | Nicolls, Fred |
| author_browse | Bradshaw, Charles Nicolls, Fred |
| author_facet | Nicolls, Fred Bradshaw, Charles |
| author_sort | Bradshaw, Charles |
| collection | Thesis |
| description | Adaptive visual object tracking (VOT) is one of the fundamental tasks in machine vision, with active research and far-reaching implications. Bayesian methods are commonly used in adaptive VOT. However, we propose that the current tendency is to restrict the inference to a subtask (e.g. classification), rather than phrasing the entire task, including the adaptive observation model, within the Bayesian inference. In this thesis we develop a framework for simultaneous modelling and estimation (SMAE), in which the common Bayesian recursive estimator (BRE) is extended to include estimation of the underlying hidden Markov model (HMM). The framework is developed not only for the task of adaptive VOT, but also for persistent tracking: the long-term task including automatic detection and tracking of multiple targets in a scene in a manner such that performance improves as a function of deployment time. To prove that the framework is usable and leads to tractable implementations, it is applied to the challenging task of maritime surveillance. Oceans provide a non-trivial noisy background against which many adaptive trackers struggle. Our developed adaptive tracker creates a baseline in which the joint distribution across observation model and target state is maintained in an adapted particle filter. A persistent tracker is then built around the adaptive tracker to produce improved results using the information from previous observations. Both the adaptive tracker and the persistent tracker use the holistic Bayesian framework described by SMAE. We find that SMAE does lead to tractable solutions that include the strength of Bayesian methods for the observation model component in adaptive VOT. In addition to this, contributions are made to the current maritime surveillance literature, in the form of a better performing salience filter for maritime and littoral scenes, and a Bayesian means for combining different salience filters. This last contribution may seem trivial; however, we were unable to find it in the maritime literature. This work also includes the application of SMAE to more philosophical topics. Although the discussion may seem informal in light of the technical nature of the body of our work, it was an integral part of the development of the framework. |
| format | Thesis |
| id | oai:open.uct.ac.za:11427/27871 |
| institution | University of Cape Town (South Africa) |
| language | eng |
| last_indexed | 2026-06-10T12:46:28.872Z |
| license_str | Not specified — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from UCTD — University of Cape Town Open Access Repository |
| publishDate | 2018 |
| publishDateRange | 2018 |
| publishDateSort | 2018 |
| publisher | Department of Electrical Engineering |
| publisherStr | Department of Electrical Engineering |
| record_format | dspace |
| source_str | UCTD — University of Cape Town Open Access Repository |
| spelling | oai:open.uct.ac.za:11427/27871 Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance Bradshaw, Charles Nicolls, Fred De Jager, Gerhard Electrical Engineering Adaptive visual object tracking (VOT) is one of the fundamental tasks in machine vision, with active research and far-reaching implications. Bayesian methods are commonly used in adaptive VOT. However, we propose that the current tendency is to restrict the inference to a subtask (e.g. classification), rather than phrasing the entire task, including the adaptive observation model, within the Bayesian inference. In this thesis we develop a framework for simultaneous modelling and estimation (SMAE), in which the common Bayesian recursive estimator (BRE) is extended to include estimation of the underlying hidden Markov model (HMM). The framework is developed not only for the task of adaptive VOT, but also for persistent tracking: the long-term task including automatic detection and tracking of multiple targets in a scene in a manner such that performance improves as a function of deployment time. To prove that the framework is usable and leads to tractable implementations, it is applied to the challenging task of maritime surveillance. Oceans provide a non-trivial noisy background against which many adaptive trackers struggle. Our developed adaptive tracker creates a baseline in which the joint distribution across observation model and target state is maintained in an adapted particle filter. A persistent tracker is then built around the adaptive tracker to produce improved results using the information from previous observations. Both the adaptive tracker and the persistent tracker use the holistic Bayesian framework described by SMAE. We find that SMAE does lead to tractable solutions that include the strength of Bayesian methods for the observation model component in adaptive VOT. In addition to this, contributions are made to the current maritime surveillance literature, in the form of a better performing salience filter for maritime and littoral scenes, and a Bayesian means for combining different salience filters. This last contribution may seem trivial; however, we were unable to find it in the maritime literature. This work also includes the application of SMAE to more philosophical topics. Although the discussion may seem informal in light of the technical nature of the body of our work, it was an integral part of the development of the framework. 2018-05-03T12:17:53Z 2018-05-03T12:17:53Z 2018 Doctoral Thesis Doctoral PhD http://hdl.handle.net/11427/27871 eng application/pdf Department of Electrical Engineering Faculty of Engineering and the Built Environment University of Cape Town |
| spellingShingle | Electrical Engineering Bradshaw, Charles Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| thesis_degree_str | Doctoral |
| title | Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| title_full | Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| title_fullStr | Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| title_full_unstemmed | Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| title_short | Structure incorporation of model uncertainty for Bayesian adaptive tracking and its application to maritime surveillance |
| title_sort | structure incorporation of model uncertainty for bayesian adaptive tracking and its application to maritime surveillance |
| topic | Electrical Engineering |
| url | http://hdl.handle.net/11427/27871 |
| work_keys_str_mv | AT bradshawcharles structureincorporationofmodeluncertaintyforbayesianadaptivetrackinganditsapplicationtomaritimesurveillance |