거리 영상과 시멘틱 세그먼테이션을 활용한 보행환경 평가 지표 개발

Abstract

The pedestrian environment is a very important element in the daily life of citizens as both an individual domain and a public space. As the importance of the pedestrian environment has been recognized, a systematic survey is needed at the national level, such as stipulating by law that local governments across the country conduct a survey of the pedestrian environment every five years. However, the survey on the pedestrian environment does not deviate from the existing limitations in the survey methodology, such as relying on field surveys in some areas. This study aimed to develop a walkability evaluation index using high-resolution streetview images and deep learning technology. To develop a walkability evaluation index, a draft of the walkability evaluation index was developed based on a review of domestic and foreign literature and a study on the evaluation of the walkability using deep learning technology. In order to confirm the possibility of constructing the derived walkability evaluation index, the final index was proposed after examining the accuracy of the result of semantic segmentation of streetview images and the possibility of obtaining necessary data. As for the derived walkability evaluation indicators, it was suggested to use 8 indicators in 4 categories: safety, convenience, comfort, and accessibility. The results of this study break away from the limitations of existing walkability studies based on field observation surveys and surveys, provide an opportunity for intelligent urban research using high-resolution streetview images and deep learning technology, and perform pedestrian environment evaluation tasks more efficiently.