Ransom, Michael ORCID: 0000-0002-4639-8099
(2024)
Comparing target tracking algorithms.
PhD thesis, University of Liverpool.
Text
201305224_Jun2024.pdf - Author Accepted Manuscript Download (9MB) | Preview |
Abstract
This thesis concerns the comparison of target tracking algorithms. A surveillance scenario may include intermittently visible targets that may be detectable by multiple sensors. Additional detections may also originate from other objects of no interest and false alarms may also be present. Quantifying the degree to which a target is detectable and trackable is a challenge that requires a solution. Models can be used to characterise a surveillance scenario. These models can also be used to simulate data. The same models can then inform development of a target tracking algorithm. A target tracking algorithm is recursive. As such, it conducts a time-evolving statistical analysis. At each iteration, an algorithm compares a prediction calculated from a previous estimate and the target tracking models with sensor observations. This comparison is used to update the estimate. Multiple competing algorithms exist. A performance quantification method quantifies the differences between a surveillance scenario realisation and an estimate from a target tracking algorithm. Comparing such algorithms would make it possible to identify the best choice of target tracking algorithm solution for a specific scenario. Many such scenarios exist. This includes scenarios with a single active sensor as well as collocated active and passive sensors. This thesis places particular emphasis on a systematic analysis investigating such comparisons. Both baseline target tracking algorithm implementations and novel developments improving on this baseline are compared. These implementations are based on a choice of algorithm for state estimation. The Kalman filter (KF) and the particle filter (PF) are two such implementations that feature herein. State estimation forms the foundation of joint target-detection and tracking (JoTT) algorithms. Two examples include the integrated probabilistic data association filter (IPDAF) and the integrated expected likelihood particle filter (IELPF). The comparisons herein consider a range of scenarios varying in complexity. The effectiveness of using multiple sensors for a target tracking solution is also evaluated. The conclusion of this thesis is that the optimal choice between either a KF or a PF is dependent on the surveillance scenario. The KF algorithm is preferable in a simple scenario (i.e., targets with a high probability of detection in low clutter and a low tolerance on availability of computational resources). Using multiple sensors results in improvements on this baseline algorithm when using sophisticated PF algorithms. Differences in relative performance are especially evident in challenging scenarios featuring dense clutter.
Item Type: | Thesis (PhD) |
---|---|
Additional Information: | Questions about this thesis can be sent to the author using the email address below. mistermransom@gmail.com |
Depositing User: | Symplectic Admin |
Date Deposited: | 01 Oct 2024 10:22 |
Last Modified: | 08 Feb 2025 03:04 |
DOI: | 10.17638/03182841 |
Supervisors: |
|
URI: | https://livrepository.liverpool.ac.uk/id/eprint/3182841 |