Examining the Occurrence Risk of Theft From Vehicle in Kyoto City, Japan Using Streetscape-level Indicators

• In Event: Advances in Spatial Methods for the Study of Crime & Place

Abstract

Objective: With the recent advances in the collection of streetscape images, new environmental indicators at the street level become available through deep learning. This study examined the environmental configurations at the street level that contribute to the risk of theft from vehicles (TFV) on streets through the new indicators and their interaction effects. Method: A binomial logistic model was conducted to predict the occurrence of TFV at sample points on the streets in Kyoto City, Japan (N=236). We used streetscape indices obtained from semantic segmentation of Google Street View images, street network centralities, and their interaction terms as explanatory variables. Results: The results revealed the complex effects of the streetscape environment on the TFV risk, which may both increase and decrease the risk, depending on other factors. For example, the association of the risk with the amount of vegetation in an image was estimated to change from negative to positive as the number of buildings and walls in the image increased. Conclusions: This study demonstrated that the streetscape indicators allow quantitative analysis of situational crime risk associated with the configuration of the streetscape environment.

Authors

• Hiroki Adachi, Graduate School of Environmental Studies, Tohoku University

• Tomoki Nakaya, Graduate School of Environmental Studies, Tohoku University