PHOEBE partner NTUA recently published a paper in the interdisciplinary journal Transportation Research Interdisciplinary Perspectives, which is concerned with social science aspects of transportation. The paper presents an integrated framework for pedestrian-vehicle tracking and behaviour analysis in urban environments using computer vision and street-level video data from smartphone cameras.
What is the aim of the study?
The study aims to enhance road safety by leveraging Surrogate Safety Measures (SSMs), which estimate the likelihood and time remaining before a collision occurs between a pedestrian and a vehicle. The system used state-of-the-art object detection and feature extraction models, such as YOLOv8 (a deep learning-based model for real-time object detection) and ResNet-50 (a convolutional neural network for feature extraction). This method is particularly useful for identifying high-risk locations and evaluating pedestrian compliance with traffic signals. By automating the detections and analysis of pedestrian movements and vehicles trajectories across multiple frames, the study’s contributes to data-driven decision making for city planners, traffic and road safety engineers and policymakers.
What data is used?
The study is based on street-level video recordings collected from smartphone cameras positioned at ground level in key locations in Athens, Greece. Unlike traditional traffic monitoring systems that rely on fixed, high-mounted cameras or expensive sensor-based equipment, this approach emphasises affordability, high flexibility and ease of deployment in different environments.
What is the accuracy rate of the system?
The system achieved an accuracy rate of 50% to 70% in detecting the status of pedestrian traffic lights, which is crucial for understanding whether pedestrians comply with signals. Moreover, the framework identified a 23% discrepancy on average between manual and automated counts of illegal pedestrian crossings, highlighting potential limitations in human observation methods. Despite these promising results, the study acknowledges challenges in real-world applications, such as occlusions, varying lighting conditions, and camera placement angles, which can impact detection accuracy.
How were smartphone cameras used?
One of the most significant contributions of this study is its demonstration of how smartphone cameras can be effectively used for large-scale, cost-efficient traffic analysis. Unlike conventional surveillance methods that require fixed, high-cost infrastructure, the proposed framework allows researchers and urban planners to deploy monitoring systems flexibly and at a lower cost. This accessibility makes it easier for city authorities to collect real-time data on pedestrian and vehicle interactions and to use this information for policy-making and infrastructure improvements.
What is the conclusion?
The study highlights the potential of computer vision in creating safer urban environments by providing a scalable and transferable method for tracking road users and improving traffic management strategies. Key points include:
- Reliable traffic data can be provided by smartphone cameras, reducing dependence on costly surveillance infrastructure.
- Automated pedestrian tracking can improve traffic management strategies and enhance road safety assessments.
- Machine learning-based detection models can help identify high-risk zones and areas where pedestrian compliance is low.