Team:

Jakson Paterson, Élyse Comeau, Alison Novak, Tilak Dutta

Current design features and maintenance approaches used at urban intersections can negatively impact pedestrian safety at street crossings, particularly for pedestrians with disabilities. Most municipalities in Canada rely on counts of collisions, injuries or fatalities at a particular intersection to indicate there may be a safety concern that needs to be addressed. Recent advanced in computer vision offer the potential to identify near misses to highlight systemic safety concerns before someone is hurt.

The objective of this work is to investigate the potential for a computer vision system to highlight pedestrian safety concerns at urban intersections. Specifically, this project investigates whether our Pedestrian Safety Evaluation system (PSE) can be used to create a new foundational framework for quantitatively evaluating pedestrian safety at urban intersections by identifying near misses and unsafe conditions.

Problem:

Pedestrians (particularly older adults and people with disabilities) can face safety and usability challenges at street crossings because of the inherent risks associated with environmental conditions (accumulations of rain, snow, ice, or poor lighting) and through their interactions with cyclists and vehicles. These external factors have both physical and psychological consequences on the perceived safety and risk of fall and injury. As a result, older adults and people with disabilities may choose to reduce their outdoor activity levels, isolating themselves in their homes, and potentially causing a downward spiral of deconditioning, frailty, and physical decline. Weakened physical strength can then increases their risk of falling and sustaining an injury in the future, which would subsequently perpetuating the fear that keeps them inside and continue this cycle of fear and limited movement in the winter. The development of more pedestrian-centric infrastructure can help to promote active transportation and reduce the winter isolation that so many older adults and persons with disabilities experience.

Currently, there is a lack of quantitative observational methods for pedestrians which has limited the understanding of the impacts that environmental barriers and other road users have on older adults and persons with disabilities. This project involves the development and evaluation of computer vision tools to investigate pedestrian safety and usability challenges at intersections. These methods will highlight limitations in current infrastructure design and maintenance so that elements of the street crossing can be made safer and more usable for all.

Goal:

The objective of this project is to automate the assessment of pedestrian behaviour at street crossings in urban areas to help understand the challenges that pedestrians face, especially individuals with the greatest safety or usability concerns. Computer vision approaches including the You Only Look Once (YOLOv7) and Simple, Online, and Real-time Tracking (DeepSORT) will be used to automate the detection and measurement of pedestrian walking behaviour, as well as the classification of conflicts between pedestrians and other road users to achieve the following aims:

1. Develop a portable camera system for recording video and integrate with YOLOv7 and DeepSORT.
2. Demonstrate the ability of computer vision techniques to identify weather-related pedestrian usability challenges.
3. Identify interactions and conflicts between pedestrians and other road users at street crossings.
4. Propose a method for assessing the accessibility of street crossing for pedestrians to compare the infrastructure of intersections within the City of Toronto. This intersection safety scoring system will work with urban planning experts in industry and academia to weigh the impacts of external risk factors on pedestrian movement.

This project will provide new automated observation tools (a) for evaluating the safety and usability of street crossings and (b) to understand the complex relationship between pedestrians and the built environment. We hope to encourage the development of solutions that promote active mobility, allowing older adults and persons with disabilities to access their daily necessities, and fulfill their sense of autonomy, and social connection to your surrounding community.

Results:

Environmental Risks

Winter weather was found to restrict the speed of pedestrians and increased the number of pedestrians (16%) who fell below the 1.0 m/s walking speed standard that most municipal signal timing assumes.

Over nine hours of video data, 3114 pedestrians were observed under clear and snow-covered road conditions and between daylight and reduced lighting. Mean pedestrian speed decreased from 1.51 ± 0.50 m/s on clear roads to 1.27±0.37 m/s under snowy conditions. Among more vulnerable populations, average speeds of the bottom fifth percentile declined from 0.72 m/s to 0.63 m/s, and the proportion of pedestrians walking below the standard 1.0 m/s crossing threshold increased from 10% to 17%.

Non-parametric analyses identified significant differences in pedestrian speed across weather and lighting conditions. Pedestrian path deviations decreased under snow and increased under reduced lighting, reflecting increased uncertainty.

Pedestrian Intersection Detection

The PSE system is able to detect conflicts between pedestrians and other road users (F1 = 0.93, sensitivity = 0.82) and highlighted the potential for identifying near-misses to supplement the existing collision-based intersection safety metrics.

Significance

The integration of computer vision analyses with accurate, large-scale behavioural data can provide municipal planners with quantitative evidence to support data-driven accessibility improvements. Winter weather is a significant barrier to pedestrian mobility and should be prioritized to restructure intersection design to support vulnerable pedestrians. The use of computer vision systems provides a data foundation for the complex decisions that connect numerous municipal departments (traffic engineering, public transportation, snow maintenance, etc.). The capabilities of this system can offer a cost-effective tool for intersection monitoring, capturing design and maintenance problems before they result in injury or fatality. The processing of hours of video data to synthesize observation and automatically flag concerning events would be a substantial advancement over manual observation methods which is resource intensive and limited in scope.

Contributors

Hamed Ghomashchi, Zeyad Ghulam, Niroja Balakumar, Thomas Elliott