Analysing gender differences in the perceived safety from street view imagery

The relationship between the built environment and human perception of safety is well recognised in a growing literature of urban studies. However, there is a lack of attention to gender differences in perceptions of place, particularly in studies that assess perceived safety using street view images (SVIs). This limitation hinders the comprehensive assessment of safety perceptions. Traditional analyses that combine gender or focus on men do not adequately address women's specific needs to feel safe. To rectify this, the 60 participants were divided into two groups based on gender. Their perceived safety scores on 1,034 SVIs, and we used regression analysis to infer similarities and differences in streetscape elements that influence the safety scores between genders. Secondly, a machine learning model was trained, considering approximately thirty streetscape elements, and used to predict the safety scores of SVIs in the city. Finally, the spatial distribution of perceived differences between genders was visualised, and portraits of the different scenes were depicted. The results show that 1) both genders' safety scores are mainly influenced by elements such as "Road", "Sidewalk", and "Car", while the impact of "Bridge" varied between genders. 2) A high correlation was observed between the predicted safety scores for women and men. However, women deemed 63% of scenes unsafe, compared to men who considered only 23% of scenes unsafe, indicating a 40% difference. 3) The safer the scene is, the smaller the difference in perception between genders. Conversely, the more unsafe the scene, the weaker women's perceptions of safety are compared to men's. Our findings can extend the rules of urban safety assessment (serving women) and create an inclusive urban street environment.