Vulnerable road users (VRUs) and their safety are a vital pillar of the Safe System approach and achieving Vision Zero objectives. This research examines pedestrian behaviour and violation patterns at both signalised and non-signalised intersections in the city centre of Athens, Greece, using video data recordings captured on mounted tripods with mobile phone cameras and computer vision algorithms as part of the PHOEBE project. The objective is to quantify compliance levels and identify the environmental, traffic, and behavioural factors influencing pedestrian crossing decisions. Video data were collected from two intersections in the center of Athens representing diverse traffic conditions, geometric layouts, and pedestrian demand levels. Using advanced computer vision and machine learning methods, pedestrian and vehicle movements were automatically detected, classified, and tracked. The analytical framework combined YOLOv8 for object detection, a Kalman filter for trajectory tracking, and a homography transformation for real world coordinates. The resulting dataset included detailed temporal and spatial features such as crossing duration, traffic light phase for the signalized crossing, pedestrian speed, and vehicle proximity. Each crossing event was automatically labelled as compliant or non-compliant according to the prevailing traffic signal and safety conditions. Subsequent statistical and data-driven modelling was conducted to determine the predictors of non-compliance. Logistic regression and Poisson regression models were applied to evaluate the influence of signal phase, vehicle flow, traffic speed, crossing width, and pedestrian characteristics on the likelihood and frequency of violations. The results reveal that compliance rates are significantly higher at signalised crossings compared with non-signalised ones. However, even at signalised locations, nearly one-third of pedestrians crossed during the red phase, particularly when perceived waiting times were long or vehicle approach speeds were low. In the case of the unsignalised crossings, violations were more frequent in smaller road segments and under conditions of moderate traffic flow or congested traffic flow, suggesting a behavioural trade-off between perceived risk and delay. The results of the study offer practical insights for urban planners and road safety authorities. Some recommended actions include improving pedestrian visibility, lowering vehicle speeds near intersections, or optimizing signal timings. The study also demonstrates how automated video analytics can provide cost-effective, scalable data for evaluating pedestrian safety and monitoring compliance in real urban environments.
| ID | pc680 |
| Presentation | |
| Full Text | |
| Tags |
