Prediction of Visual Quality of Streetscape Images Using Computer Vision

An unpleasant urban environment often discourages physical activity, but creating a pleasant environment can improve physical and mental health. Moreover, micro-scale environmental components that can be modified cost-effectively can markedly impact walking. This research proposed and implemented a framework to predict visual pleasantness-unpleasantness and assessed the effect of each component in streetscape images based on micro-scale environmental components through computer vision. The framework achieved an overall F1 score of 0.946 in detecting micro-scale components and materials via computer vision. Pleasantness and unpleasantness were predicted with an accuracy of 0.809 and 0.743, and the presence of green spaces, cars, trucks, curbs, and roads, and materials of sidewalks, fences, and walls had a significant effect on the pleasantness-unpleasantness. This study contributes to urban environment evaluation by proposing a method to predict the pleasantness-unpleasantness of streetscapes and is expected to practically improve the urban environment at the street level.