Detecting Ill-intended Crossing Behaviour of Pedestrians
Keywords
Loading...
Authors
Issue Date
2022-10-12
Language
en
Document type
Journal Title
Journal ISSN
Volume Title
Publisher
Title
ISSN
Volume
Issue
Startpage
Endpage
DOI
Abstract
This thesis aims to improve future urban traffic by focusing on interactions between Autonomous
Vehicles (AVs) and Vulnerable Road Users (VRUs) in crossing scenarios. The main
research question is “To what extent can ill-intended crossing behaviour of pedestrians be recognized
in a morally responsible way, in the context of AVs?” Specifically, the aim is to avoid a
Pedestrian Supremacy, the scenario where VRUs are crossing the road illegally, taking advantage
of the safety-first AVs, thereby stalling traffic. This scenario might work against the transition towards
AVs, which is undesirable as AVs promise a safer traffic when compared to Human-Driven
Vehicles (HDVs). In the experiment the Pedestrian Intention Estimation (PIE) dataset, containing
video data, is decomposed into categorical, textual data. Binary Logistic Regression (BLR) is
used, evaluating the resulting categorical data. It was found that the PIE dataset might suffice for
the task, suggesting that there is a potential for a method that can evaluate intentions. Yielding
high accuracy of 0.9444 using the sex, action and age predictors, analysed with an inexpensive
method such as BLR shows that the method is able to run in real time. Further optimisation and
more complex models could improve classification performance. In the current form, improvements
to the dataset are necessary to avoid bias. Suggestions to improve the dataset as well as
suggestions for further research are made. Special attention is attributed to safety, privacy, lawfullness
and moral responsibility. The goal is to lay the foundation for an ethically sound agent
that is aiding the transition towards safer, autonomous traffic.
Description
Citation
Supervisor
Faculty
Faculteit der Sociale Wetenschappen