Automated Assessment of Pedestrian-Safety Features from Street-View Imagery Using Semantic Segmentation and Self-Supervised Representation Learning: A Citywide Study in Isfahan

Document Type : Original Article

Authors
1 Department of Civil Engineering, Ki.C., Islamic Azad University, Kish, Iran
2 Associate Professor, Transportation Research Institute, Road, Housing and Urban Development Research Center
3 Assistant Professor,Department of Civil Engineering, SR.C., Islamic Azad University, Tehran, Iran
Abstract
Microscale characteristics of the built environment substantially influence pedestrian safety; however, conventional field audits are costly, subjective, and difficult to implement at the city scale. This study develops and validates a hybrid computer-vision framework for the automated assessment of pedestrian safety using street-view imagery. The framework integrates semantic segmentation for extracting explicit and interpretable environmental attributes with self-supervised representation learning for capturing latent visual characteristics, including pavement condition, spatial enclosure, material quality, and visual disorder. The empirical analysis covers 368,500 street-view images collected from 92,125 georeferenced locations in Isfahan, Iran. A fine-tuned SegFormer-B4 model was used to quantify eight safety-related features, while a DINO-trained Vision Transformer generated contextual visual representations. These feature streams were fused within an XGBoost regression model to construct the Automated Pedestrian Safety and Environment Index (APSEI), using expert field-audit scores from 500 stratified locations as ground truth. External validity was evaluated against 3,212 geocoded pedestrian–vehicle crashes through geographically weighted Poisson regression.
Keywords

 Arellana J, Saltarin M, Larranaga AM, et al. (2020). Walkability indicators and their association with pedestrian crashes. Transp Res Part C. 111:156-175.
-­Bugs G, Granell C, Fonts O, et al. (2010). An assessment of Public Participation GIS. Cities. 27(3):172-181.
- Caron M, Touvron H, Misra I, et al. (2021). Emerging properties in self-supervised vision transformers. IEEE ICCV; 9630-9640.
- Chen S, Kuhn M, Prettner K, Bloom DE. (2019). The global economic burden of road traffic injuries. Lancet Glob Health. 7(12):e1663-e1674.
-­Hadjidemetriou GM, Vela PA, (2020). Christodoulou SE. Vision-based deep learning for automated detection of pedestrian-vehicle conflicts. Accid Anal Prev. 135:105355.
- Hall CM, Ram Y. (2018). Walk Score and its potential contribution. Transp Res Part D. 61:310-324.
- Jacobsen PL. (2003). Safety in numbers. Inj Prev. 9(3):205-209.
- Millington C, Ward Thompson C, Rowe DA, et al. (2009). Development of the Scottish Walkability Assessment Tool. Health Place. 15(2):474-481.
-­Nasar JL. (2008). Assessing perceptions of environments for active living. Am J Prev Med. 34(4):357-363.
- Yin L, Cheng Q. Wang Z, Shao Z. (2015). 'Big data' for pedestrian volume. Comput Environ Urban Syst. 53:96-108.
- Yin L, Wang Z. (2016). Measuring visual enclosure for street walkability. Comput Environ Urban Syst. 59:164-175.
- Zhang F, Zhou B, Liu L, et al. (2018). Measuring human perceptions. Landsc Urban Plan. 180:148-160.
-Aghaabbasi M, Moeinaddini M, Zaly Shah M, Asadi-Shekari Z. (2020). A new assessment model. Accid Anal Prev. 134:105336.
-Ayush K, Uzkent B, Meng C, et al. (2021). Geography-aware self-supervised learning. IEEE ICCV; 10181-10190.
-Biljecki F, Ito K. (2021). Street view imagery in urban analytics and GIS: A review. Landsc Urban Plan. 215:104217.
-Brownson RC, Hoehner CM, Day K, et al. (2009). Measuring the built environment for physical activity. Am J Prev Med. 36(4 Suppl):S99-S123.
-Carr LJ, Dunsiger SI, Marcus BH. (2010). Walk Score™ as a global estimate. Am J Prev Med. 39(5):460-463.
-Chen LC, Zhu Y, Papandreou G, et al. (2018). Encoder-decoder with atrous separable convolution. ECCV, 801-818.
-Chen T, Kornblith S, Norouzi M, Hinton G. A (2020). simple framework for contrastive learning. ICML, 1597-1607.
-Day K, Boarnet M, Alfonzo M, Forsyth A. (2006). The Irvine-Minnesota inventory. Am J Prev Med. 30(2):144-152.
-Ewing R, Cervero R. (2010). Travel and the built environment: A meta-analysis. J Am Plann Assoc. 76(3):265-294.
-Harvey C, Aultman-Hall L. (2016). Measuring urban streetscapes for livability. Prof Geogr. 68(1):149-158.
 -He K, Fan H, Wu Y, et al. (2020). Momentum contrast for unsupervised visual representation learning. IEEE CVPR, 9729-9738.
-Ibrahim MR, Haworth J, Cheng T. (2020). Understanding cities with machine eyes. Cities. 96:102481.
-Koo B, Guhathakurta S, Botchwey N. (2022). Development and validation of a street-level pedestrian safety index. J Transp Geogr. 99:103300.
-Koo TK, Li MY. (2016). A guideline of selecting and reporting intraclass correlation coefficients. J Chiropr Med. 15(2):155-163.
-Li X, Ratti C, Seiferling I. (2018). Assessing street-level urban greenery. Urban For Urban Green. 34:52-59.
-Mooney SJ, DiMaggio CJ, Lovasi GS, et al. (2016). Use of Google Street View. Am J Public Health. 106(5):876-82.
-Naci H, Chisholm D, Baker TD. (2009). Distribution of road traffic deaths by road user group. Inj Prev. 15(1):55-9.
-Naik N, Philipoom J, Raskar R, Hidalgo C. (2014). Streetscore. IEEE CVPR Workshops, 779-786.
-Rundle AG, Bader MDM, Richards CA, et al. (2011). Using Google Street View to audit neighborhood environments. Am J Prev Med. 40(1):94-100.
-Sleet DA, Pollack Porter KM, Ewing R, et al. (2020). Transportation and Public Health. Elsevier.
-Stoker P, Garfinkel-Castro A, Khayesi M, et al. (2015). Pedestrian safety and the built environment. J Transp Land Use. 8(2):81-105.
-World Health Organization (2018). Global status report on road safety Geneva.
-Zhao H, Shi J, Qi X, et al. (2017). Pyramid scene parsing network. IEEE CVPR, 2881-2890.
-Zuniga-Teran AA, Orr BJ, Gimblett RH, et al. (2017). Designing healthy communities. Front Archit Res. 6(1):63-73.