Myeonggi Hong, Eunbi Cho, Kwanghoon Pio Kim, Journal of Internet Computing and Services, Vol. 25, No. 6, pp. 37-48, Dec. 2024
10.7472/jksii.2024.25.6.37, Full Text:  HTML
Keywords: Environmental criminology, Sexual crime, Cityscape, Semantic segmentation, Convergence

Abstract

This study analyzes the spatio-temporal patterns of street-level sexual crimes and explores the micro-environmental factors associated with crime locations using deep learning-based Semantic Segmentation techniques. We examined 122 sexual crime incidents that occurred in specific area, from 2010 to 2015, alongside Kakao Street View images to assess the environmental characteristics of crime scenes. Our findings reveal a significant temporal distribution, with higher crime rates observed during evening and nighttime. Hotspot analysis utilizing Getis-Ord G* indicates that evening crimes predominantly cluster in commercial areas, whereas nighttime incidents are more concentrated in residential zones. Additionally, the results from PSPNet semantic segmentation indicate that crime locations consistently exhibit a high pixel ratio of buildings and a low pixel ratio of sky. Notably, variations in pixel ratios of key classes were observed across different time periods. This research underscores the application of computer engineering technologies in criminological studies and suggests potential foundational insights for future crime prevention strategies.

Statistics
Cite this article