Using Google Street View imagery to capture micro built environment characteristics in drug places, compared with street robbery

Abstract

The drug-related problem poses a serious threat to human health and safety. Previous studies have associated drug places with factors related to place management and accessibility, often at several scattered places, as data at the micro level are hard to obtain at a city-wide scale. Google Street View imagery presents a new source for deriving micro built environment characteristics, including place management and accessibility in larger areas. In this study, we calculate an overall safety score by the Streetscore algorithm and extract physical elements at the address location by the Pyramid Scene Parsing Network (PSPNet) model from every Google Street View image. Additionally, to distinguish drug activities from other types of crime, we compare drug-related calls for service (CFS) data with street robbery incident data. We build the binary logistic regression models to assess the impact of the micro built environment variables on drug activities after controlling for other criminological elements pertaining to drug places. Results show that the safety score, traffic lights, and poles make statistically significant and negative (or deterring) impacts on drug activities, whilst traffic signs and roads make statistically significant and positive (or contributing) impacts. The positive impact of buildings is also notable as its p-value is slightly over 0.05. This study provides evidence at the micro level that less place management and higher accessibility can increase the risk of drug activities. These street-view variables may be generally applicable to other types of crime research in the context of the micro built environment.