Modelling the impact of sensor placement and fusion for traffic monitoring

Download files
Access & Terms of Use
open access
Copyright: Laird, John
Altmetric
Abstract
This thesis develops models to evaluate the impact of multi-modal sensor placement configurations on obtaining traffic parameters required for a variety of traffic monitoring and management applications. Existing traffic management strategies generally rely on the commonly used induction loop sensors, which are highly accurate presence detectors, however they have a limited sensing area. Alternate sensor modalities may provide a higher information gain in comparison, especially vision based sensors. An inherent problem with using vision based sensors in traffic management is the occlusion between vehicles, which can make detection of individual vehicles difficult. Information fusion from multiple sensors provides much richer information for scene understanding, leading to a greater ability to coordinate traffic management efficiently. Thus, effective sensor placement and fusion of data can improve the efficiency of traffic management. The aim of this thesis is to evaluate the impact of multi-modal sensor placement, and as a result improve the estimation accuracy of road traffic parameters obtained from various sensor configurations. More specifically, vision based sensors are studied, and later fused with inductive loops. In order to achieve this, models are developed to simulate various traffic flows, and to ensure consistency and relevance of the simulations to real world traffic. Models for single sensor modality and multi-modal sensor fusion are developed and are also validated. These models enable evaluation of sensor placement configurations to determine effectiveness for specific traffic applications. Results of various single sensor configurations demonstrate that the impact of view occlusion on the ability to detect vehicles can be improved by considering sensor placement. Combining inductive loops and video cameras by sensor fusion was found to overcome the problem of occlusion, resulting in a decided improvement in parameter estimation for traffic management and monitoring applications. For example, the sensor fusion models developed resulted in the queue length estimation accuracy being improved by up to 20%. Finally, by applying the models presented in this thesis to current ramp metering strategies, viable alternative sensor deployment solutions are recommended.
Persistent link to this record
Link to Publisher Version
Link to Open Access Version
Additional Link
Author(s)
Laird, John
Supervisor(s)
Chou, Chun Tung
Geers, D. Glenn
Wang, Yang
Creator(s)
Editor(s)
Translator(s)
Curator(s)
Designer(s)
Arranger(s)
Composer(s)
Recordist(s)
Conference Proceedings Editor(s)
Other Contributor(s)
Corporate/Industry Contributor(s)
Publication Year
2013
Resource Type
Thesis
Degree Type
PhD Doctorate
UNSW Faculty
Files
download whole.pdf 2.73 MB Adobe Portable Document Format
Related dataset(s)