거리 영상과 시멘틱 세그먼테이션을 활용한 보행환경 평가 = Assessing Walkability Using Street View Images and Semantic Segmentation Technique

다국어 초록 (Multilingual Abstract)

Walking is recognized for its positive effects in various aspects such as promoting residents' health, serving as a green mode of transportation, and strengthening community ties. As the benefits of walking become more widely acknowledged, creating a pedestrian-friendly environment has become a crucial element in urban management. Consequently, objective and quantitative evaluations of the walking environment at the street level have become a key topic. While many previous studies have identified and specified factors affecting walking through field surveys or questionnaires, these studies were mostly limited to specific areas or particular groups of residents. Recent advancements in street view images provided by Internet portals such as Google, along with developments in deep learning technology, have made it possible to closely observe the urban built environment at a detailed street level. This study aims to assess the pedestrian environment at this detailed street level by utilizing street view images and deep learning technology. To achieve this, nine evaluation indicators were established across four categories: safety, convenience, comfort, and accessibility. Semantic segmentation techniques were applied to street view images, data crawling was conducted using APIs, and GIS-based analytical methods were employed to compile a dataset of pedestrian environment evaluation indicators and conduct a comprehensive assessment of the pedestrian environment. This study is significant in that it establishes a detailed and efficient framework for evaluating the pedestrian environment at the street level using street view images and deep learning technology. Additionally, its practical applicability was demonstrated by applying the framework to analyze and propose improvements for the pedestrian environment in the study area of Anyang City.