Automated Multiview Safety Analysis at Complex Road Intersections
| aut.embargo | No | en_NZ |
| aut.thirdpc.contains | No | en_NZ |
| dc.contributor.advisor | Klette, Reinhard | |
| dc.contributor.advisor | Chien, Hsiang-Jen (Johnny) | |
| dc.contributor.author | Moayed, Zahra | |
| dc.date.accessioned | 2020-06-25T22:56:22Z | |
| dc.date.available | 2020-06-25T22:56:22Z | |
| dc.date.copyright | 2020 | |
| dc.date.issued | 2020 | |
| dc.date.updated | 2020-06-25T08:50:37Z | |
| dc.description.abstract | The safety of pedestrians and vehicles at traffic intersections is a major concern for transport practitioners these days due to the high number of reported accidents and fatalities. Computer vision as an essential part involved in intelligent transport systems that can take advantage of infrastructure-based recordings, such as surveillance cameras to assess events and analyse participants’ safety. Previous studies have revealed that the safety factors are investigated solely, and there is a demand to have an automated safety analyser, which considers the interaction among all participants at intersections. Due to variations in traffic scenes in terms of weather conditions and time of day, further research is still needed to achieve robustness. Most monitoring systems are designed to work in controlled environments, so the analysis might not be a good sample of real traffic intersections. Furthermore, the restricted camera view leads to having an incomplete analysis. In order to resolve these issues, an automatic vision-based system is proposed that is used to understand traffic patterns and to analyse participants’ safety at intersections. The major novelty of this thesis is to present a robust safety analyser using four calibrated cameras at a real intersection. Understanding object locations in world coordinates from different cameras helps to improve the tracking and to address the occlusion problem. Also, it yields a larger field of view for covering more areas. To inspect safety, the characteristics of the participants are extracted; detection, classification, and tracking use a fusion of appearance-based and motion-based methods. Deep learning proves its ability to take part at this stage, handling tradeoffs among accuracy, time sufficiency, and robustness, while being associated with motion parameters. The study is further extended to consider the past and future movements, together with safety measurements and interaction risk factors to analyse the potential risks for each participant in the form of a single value. As a result of this study, some attributes and distributions can be extracted for deriving an understanding of the road intersection for further design and planning to mitigate traffic risks. This research will provide a more cost-effective and reliable approach for informing participants about possible risks from the infrastructure side due to the low cost of the cameras, hence it also aligns with future technologies such as autonomous vehicles. Regarding showing effectiveness and robustness in practice, a busy road intersection at Auckland, New Zealand, is selected as the ultimate goal for monitoring. | en_NZ |
| dc.identifier.uri | https://hdl.handle.net/10292/13450 | |
| dc.language.iso | en | en_NZ |
| dc.publisher | Auckland University of Technology | |
| dc.rights.accessrights | OpenAccess | |
| dc.subject | Multiview Analysis | en_NZ |
| dc.subject | Camera Calibration | en_NZ |
| dc.subject | Traffic Safety | en_NZ |
| dc.subject | Intersections | en_NZ |
| dc.subject | Depp Learning | en_NZ |
| dc.title | Automated Multiview Safety Analysis at Complex Road Intersections | en_NZ |
| dc.type | Thesis | en_NZ |
| thesis.degree.grantor | Auckland University of Technology | |
| thesis.degree.level | Doctoral Theses | |
| thesis.degree.name | Doctor of Philosophy | en_NZ |