| Abstract |
Roundabouts are increasingly used across Canada to enhance vehicular safety and operational efficiency; however, they remain difficult for pedestrians with vision loss (PWVL) due to the lack of pedestrian signalization, irregular geometry, and continuous vehicle flow. These challenges highlight a critical accessibility gap in modern intersection design. As part of an ongoing research program at Lakehead University, this study builds on previous phases that developed and field-tested an infrastructure-mounted, sensor-agnostic accessibility system designed to assist PWVL in identifying safe crossing opportunities for roundabouts. The system integrates smart-camera detection and real-time feedback to promote independent, informed pedestrian decision-making.
This paper focuses on the data-analysis phase of the project, which aims to evaluate system performance using field data collected from a roundabout in Thunder Bay, Ontario. The analysis seeks to determine how accurately the system identifies safe crossing gaps, models pedestrian accessibility performance, and measures safety reliability under real-world operating conditions.
Field data has been collected using a smart camera mounted on a pole. The device monitored vehicle trajectories, approach speeds, and crossing events while the system provided real-time auditory or haptic feedback to test participants. The dataset includes vehicle time-to-arrival (TTA), pedestrian crossing times (TPED), and feedback response timestamps.
Analytical procedures include Poisson regression to model gap and vehicle arrival frequency, and logistic regression to classify crossing outcomes (“safe” vs. “unsafe”) based on computed TTA. Performance is assessed through detection accuracy, false-alarm rate, and latency between detection and pedestrian feedback. Environmental variables, such as lighting, weather, and approach geometry, are incorporated to evaluate the robustness of the followed algorithm.
Preliminary checks indicate stable vehicle detection, consistent TTA estimation, and low communication latency. Statistical results quantify algorithm reliability and identify operational thresholds for accurate, real-time feedback. These findings suggest a promising accessibility system that can enhance the independent mobility and confidence of PWVL. This analysis phase bridges laboratory development and large-scale validation of a novel accessibility system for unsignalized roundabouts. The results provide empirical evidence to refine the algorithm, guide future accessibility-oriented design, inform municipal policy decisions, and contribute to Canada’s broader vision of inclusive, technology-enabled, and human-centred mobility systems.
|
