A Multi-source Data Based Study On The Association Between Physical Environment And Perceived Safety

Perceived safety is related to the vitality of public space and people's well-being.It is crucial to understand the planning and design factors affecting perceived safety and the influential mechanism.According to routine activity theory,theories based on evolutionary psychology and broken windows theory,the physical environment can influence safety perception in various ways.However,in China,the empirical studies to test this environmentalpsychological relationship is still insufficient.Among such studies,the comprehensive examination on the three types of theories are not only lacking,but also coarse-grained – their results stay at the campus and community level,not involving smaller scale spaces.On the other hand,the popularity of open data and machine learning today has reduced the cost of largescale,high-resolution space research,but few environmental-safety perception studies conducted such new methods.Therefore,this study uses the open data to empirically examine the physical environmental factors that affect perceived safety and the influential mechanism,and test the applicability of open data in environmental-safety perception research.Reviewing existing theories,empirical findings and research methods,this study selects 90 paths in and around the Sipailou campus of Southeast University,Gulou Campus of Nanjing University,and Xikang Road Campus of Hohai University,and obtains perceived safety data through questionnaire survey(178 validated questionnaires).Also,this study establishes a set of material environment variables including accessibility,function and density,street view features,path edges and width,social control representation and other variables.This study employs spatial syntax to obtain accessibility data,uses POIs to calculate function and density data,applies semantic segmentation technology to calculate street view feature data,and acquires path interface and social control representation data from experts' assessing.Using ArcGIS and R language,data are fused and tidied.Associational analysis methods such as regression analysis and exploratory factor analysis are used.This study finds that(1)Street View image features can explain perceived safety variation(R2 is about 0.49~0.57),and large-scale data(space syntax,POI,Street View image elements)can better explain perceived safety variation(R2 is about 0.59~0.72).The association of street view image features with perceived safety can be explained by theories based on evolutionary psychology.Therefore,it is feasible to use open data to study safety perception.(2)8 environmental factors are found to affect safety perception: for all path samples,reducing building density,increasing sidewalk view,increasing lane view,and increasing lighting can improve safety perception,among factors lighting is the most influential and lane view is the second;Increasing the edge locomotive permeability can only significantly improve the safety perception of off-campus paths.Increasing the density of facilities and reducing the density of gates can only significantly improve the safety perception of on-campus paths.Narrowness is the most important factor in shaping the most unsafe path.Densities of facilities and gates are the most important factors in shaping the safest path;improving lighting has general effects.(3)In terms of influential mechanism,only routine activity theory and the theories based on evolutionary psychology are verified,and the broken windows theory has not been verified.