Evaluating the Performance of Walking Spaces Considering Passage and Retention Functions Using Image Recognition AI Model

Abstract

Recently, human-centered and local-context-oriented streets have been more focused, in addition to the conventional vehicle-oriented street smoothing traffic flow of vehicles. Especially in the new normal era, new local design is required to enhance people's sense of self-efficacy and happiness. The quality of pedestrian space from the viewpoint of walkability and lingerability needs to be improved to realize human-centered transportation machizukuri. Therefore, this study aims to evaluate the spatial performance of the streets and analyze the factors influencing walkability and lingerability by utilizing image recognition technology. An image recognition using fine-tuned AI model was conducted to evaluate five pedestrian spaces. Then, spatial elements influencing spatial performance were visualized and identified by using Grad-CAM. After that, respective pedestrian space is positioned in 2D coordinates with walkability and lingerability axes, and the spatial performances of close streets are compared. Finally, the potential contribution of the proposed model to improve the walkability and lingerability of streets was summarized.