Environment and Social Psychology

Estimating perceived street safety via pairwise comparisons and semantic segmentation with a social-psychological lens

Haoyuan Xiao

Architecture and Design, Nagoya Institute of Technology, Nagoya, Japan

Yoshinori NATSUME

Architecture and Design, Nagoya Institute of Technology, Nagoya, Japan

DOI: https://doi.org/10.59429/esp.v10i11.4236

Keywords: Urban street; perceived safety; CPTED; semantic segmentation; pairwise-comparison; transfer learning; social perception; environmental cues

---

Abstract

We integrate pairwise image comparisons with Semantic Segmentation to assess perceived street safety through a social-psychological lens. Drawing on classic findings about natural surveillance, signs of disorder, and risk appraisal, we pre‑specified simple directional expectations: brighter and cleaner scenes and those affording visibility should feel safer; visible rubbish and graffiti should depress safety appraisals; moderate human presence should increase perceived safety by signaling guardianship. Using 20 photos from the Shinsakae district (Nagoya, Japan), 69 participants completed 13,110 pairwise choices (all 190 combinations). A Mask2Former model, pretrained on ADE20K and fine-tuned on 263 locally annotated photos, improved mIoU from 34.15% to 66.10% and yielded area ratios for CPTED-relevant elements (lighting, greenery, people, cars, bicycles, rubbish, graffiti). We then estimated a weighted scoring function mapping these visual features to perceived-safety scores. The AI scores broadly tracked human rankings and reproduced expected social-psychological regularities: lighting/cleanliness associated positively with perceived safety, while rubbish/graffiti associated negatively; daylight and a sense of openness mattered across groups; gender, age, and nationality revealed interpretable differences in emphasis (e.g., women prioritized lighting; older adults weighted illumination more strongly; Japanese participants were more sensitive to cleanliness). We discuss how environmental cues shape quick, intuitive judgments of safety and how AI-assisted diagnostics can operationalize CPTED-informed improvements.

---

References

[1]. 1.Hino, K. , Koide, O . : Effect of Adopt-a-park-program with Regard to Fear of Crime in Parks, journal of Architecture and Planning, Vol. 70, No. 592, pp. 117-122, 2005. 6(in Japanese)

[2]. 2.Hino, K. , Ishii, N. , Hijikata, T. , Hino, A. , Amemiya, M. : Natural Surveillance on a Pedestrian Street in Tsukuba Science City, Japan: Utilization of the survey for the amount of “eyes on the street”, Reports of the City Planning Institute of Japan, Vol. 9, No. 2, pp. 64-68, 2010. 8(in Japanese)

[3]. 3.Jeffery, C. R.: Crime Prevention Through Environmental Design. Beverly Hills, CA: Sage Publications, 1971

[4]. 4.Newman, O. Defensible Space: Crime Prevention through Urban Design. Macmillan Publishing Company, 1973

[5]. 5.Cozens, P. M., Hillier, D., & Prescott, G.: Crime and the Design of Residential Property: Exploring the Theoretical Background, Property Management, Vol. 19, No. 4, pp. 222-248, 2001.5

[6]. 6.Cozens, P. , Love, T. : A Review and Current Status of Crime Prevention through Environmental Design (CPTED), Journal of Planning Literature, Vol. 30, No. 4, pp. 393-412, 2015.8

[7]. 7.Naik, N . , Philipoom, J. , Raskar, R. , Hidalgo, C. : Streetscore–Predicting the Perceived Safety of One Million Streetscapes, 2014 IEEE Conference on Computer Vision and Pattern Recognition Workshops, 2014.7

[8]. 8.Xiao, H. , Natsume, Y. : Research on factors affecting road environment safety perception and evaluation based on CPTED theory, 14th International Symposium on Architectural Interchanges in Asia, 2024. 9

[9]. 9.Takase, D. , Nambu, S. , Hino, K. , Tanaka, Y. : ESTIMATING ANXIETY CAUSED BY FEAR OF NIGHTTIME CRIME FROM PHYSICAL ENVIRONMENTAL CHARACTERISTICS, journal of Architecture and Planning, Vol. 87, No. 792, pp. 329-336, 2022. 2(in Japanese)

[10]. 10.Dubey, A., Naik, N., Parikh, D., Raskar, R., Hidalgo, C. : Deep Learning the City: Quantifying Urban Perception at a Global Scale, European Conference on Computer Vision (ECCV 2016), PP. 196-212, 2016.9

[11]. 11.Li, X., Zhang, C., Li, W., Ricard, R., Meng, Q., Zhang, W. : Assessing Street-Level Urban Greenery Using Google Street View and a Modified Green View Index, Urban Forestry & Urban Greening, Vol. 14, No. 3, pp. 675-685, 2015.6

[12]. 12.Ge, X., Shen, X., Zhou, Y.: DDC‑Net: Semantic Segmentation for Urban Roads Based on Improved Capsule Networks. Journal of Applied Science and Engineering (JASE), 28(10), 2163–2176, 2025. https://doi.org/10.6180/jase.202510_28(10).0009

[13]. 13.Yin, S., Wang, L., Chen, T., et al.: LKAFormer: A Lightweight Kolmogorov–Arnold Transformer Model for Image Semantic Segmentation. ACM Transactions on Intelligent Systems and Technology (TIST), 2025. https://doi.org/10.1145/3759254

[14]. 14.Wang, Y.: MRCNNAM: Mask Region Convolutional Neural Network Model Based on Attention Mechanism and Gabor Feature for Pedestrian Detection. Journal of Applied Science and Engineering, 26(11), 1555–1561, 2023. (Indexed in DOAJ.)

[15]. 15.Yin, S., Wang, L., Teng, L.: Threshold Segmentation Based on Information Fusion for Object Shadow Detection in Remote Sensing Images. Computer Science and Information Systems, 21(4), 1221–1241, 2024. https://doi.org/10.2298/CSIS231230023Y

[16]. 16.Teng, L., et al.: FLPK‑BiSeNet: Federated Learning Based on Prior Knowledge and Bilateral Segmentation Network for Image Edge Extraction. IEEE Transactions on Network and Service Management, 20(2), 1529–1542, 2023. https://doi.org/10.1109/TNSM.2023.3273991

[17]. 17.Jiang, Y.: A Novel DABU‑Net Model Based on Principal Component Analysis for Intelligent Collaborative Robot Design. Journal of Applied Science and Engineering, 27(11), 3533–3541, 2024. https://doi.org/10.6180/jase.202411_27(11).0010