Modeling the effects of weather and time of day on people's subjective evaluation of street impressions

Abstract

These days, modeling people's evaluation of street impressions has been attempted using big image data such as Google Street View and machine learning. However, the images used are mainly limited to those taken during the daytime in good weather conditions. This paper proposed quantitatively analyzing the effects of different weather conditions on people's evaluation of street impressions using street images collected while riding a bicycle. We conducted a web questionnaire using the collected images and constructed a street impression evaluation model trained by the questionnaire results. Using the model showed that it is possible to identify streets where the scores of certain items increase at night, for example. Furthermore, we performed a multiple regression analysis using the above conditions and the composition ratio of street components in each image as explanatory variables and the street impression evaluation scores as explained variables. As a result, we quantitatively verified the importance of weather and time of day in evaluating street impressions.