Crossing speed varies by country
Across included countries and territories, mean pedestrian crossing speed was 1.20 m/s. China showed the fastest observed mean speed at 1.69 m/s, while Chile showed the slowest at 0.88 m/s.
AutomotiveUI '25
CROWD National
What YouTube driving from 233 countries and territories teaches us about global pedestrian crossing behaviour.
233
Countries and territories in the source coverage
2,495
Cities represented in the CROWD footage map
4,122.28 h
Initial dashcam footage used for the study
124
Countries and territories after filtering
3,388.88 h
Filtered dashcam video analysed
1.20 m/s
Mean pedestrian crossing speed
Study summary
Crossing behaviour at global scale
CROWD National uses dashcam footage referenced by the CROWD dataset to analyse pedestrian crossing initiation time, crossing speed, and contextual variables across countries and territories. The study combines object detection and tracking with socioeconomic indicators such as traffic mortality, GDP, literacy, median age, and Gini coefficient to explore how pedestrian behaviour changes across global urban environments.
Key findings
What the analysis found
Initiation time also varies globally
The average crossing initiation time was 3.18 s. Qatar had the longest observed mean initiation time at 6.44 s, while China had the shortest at 1.61 s.
Hesitation and speed are linked
Crossing speed and crossing initiation time were negatively correlated (r = -0.18), suggesting that pedestrians who hesitate longer before entering the road tend to cross more slowly.
Socioeconomic context matters
Crossing speed was negatively correlated with Gini coefficient (r = -0.19) and positively correlated with traffic mortality (r = 0.18), pointing to broader links between road safety, inequality, and pedestrian behaviour.
Similar timings can emerge from different systems
Bangladesh and the Netherlands showed similar mean crossing initiation times, about 3.42 s and 3.40 s, despite very different infrastructure and pedestrian priority contexts.
Day and night effects are uneven
The analysis stratified behaviour by day and night where enough valid pedestrian crossings were detected, revealing country-specific differences in both speed and initiation time.
Method
How crossings were measured
Data source
CROWD dashcam footage
The study used the 24 July 2025 version of CROWD, with publicly accessible YouTube dashcam footage. Only car-based clips were retained for the pedestrian behaviour analysis.
Detection
YOLOv11x and tracking
YOLOv11x with ByteTrack was used to detect and track pedestrians and other road users. Additional trajectory filters removed likely riders and apparent crossings caused by camera movement.
Metrics
Speed and initiation
A crossing was defined as a lateral movement across the frame. Pixel distances were converted to real-world units using national average human height as a frame-level reference scale.
Limitations
How to read the results
The paper notes uneven dashcam availability across countries and territories, a focus on urban footage, uncertainty from converting pixel measurements using national average height, and limitations from YOLOv11x object classes and tracking errors. The authors frame the study as a large-scale behavioural analysis and a use case for CROWD, not as manually verified ground truth for every crossing event.
Citation
How to cite
@inproceedings{alam2025pedestrian,
title = {Pedestrian Planet: What YouTube Driving from 233 Countries and Territories Teaches Us About the World},
author = {Alam, Md Shadab and Martens, Marieke H. and Bazilinskyy, Pavlo},
booktitle = {Proceedings of the 17th International Conference on Automotive User Interfaces and Interactive Vehicular Applications},
year = {2025},
doi = {10.1145/3744333.3747827}
}