115 0

서울시 가로경관 이미지에 대한 주관적 인지에 영향을 미치는 가로환경 요인 분석 : Deep Learning 의미론적 분할과 YOLOv3 객체 검출기법을 적용하여

Title
서울시 가로경관 이미지에 대한 주관적 인지에 영향을 미치는 가로환경 요인 분석 : Deep Learning 의미론적 분할과 YOLOv3 객체 검출기법을 적용하여
Other Titles
Analysis of Street Environmental Factors Affecting Subjective Perceptions of Streetscape Image in Seoul, Korea : Application of Deep Learning Semantic Segmentation and YOLOv3 Object Detection
Author
이수기
Keywords
보행만족도; 가로경관; 딥러닝; 객체 검출; 의미론적 분할; Pedestrian Satisfaction; Streetscape; Deep Learning; Object Detection; Semantic Segmentation
Issue Date
2021-04
Publisher
대한국토·도시계획학회
Citation
국토계획, v. 56, NO. 2, Page. 79-93
Abstract
본 연구는 보행 가로환경 이미지를 활용한 설문조사를 토대로딥러닝 의미론적 분할(semantic segmentation) 기법과 객체 검출(object detection) 기법을 활용하여 이미지에 대한 보행만족도와 주관적 인지를 탐구하였다. 이를 위해 가로환경 요소를 포함하고 있는 GSV 이미지를 구득하였다. 또한, 사진 내 다양한 요소는 의미론적 분할기법과 객체 검출기법을 통해 추출하여 보행만족도를 포함한 주관적 인지와의 관계를 분석하였다. 분석 모형으로는 평가된 표본 이미지 셋(set)과 평가자인 개인의 차이를 구분할 수 있도록 다수준 순서형 로지스틱 회귀모형을 사용하였다. 더불어 보행환경에 대한 충분한 평가를 위해 전체적인 만족도외 긍정적인 감정에 해당하는 안전함(safety), 아름다움(beauty), 활기참(vitality) 그리고 부정적인 감정을 나타내는 지루함(boring), 우울함(depression)을 평가 항목에 도입하여 각가로환경 요소가 다양한 감정에 미치는 차별적인 영향력을 비교분석하였다. 나아가, 본 연구에서 활용된 딥러닝 모델이 가지는의의와 분석 결과를 토대로 보행환경 개선을 위한 정책적 시사점을 제시하였다. This study aims to demonstrate the relation between streetscape features and six different individual perceptions, which includes pedestrian satisfaction. Two deep learning techniques—semantic segmentation and object detection—were applied on Google Street View imagery; these techniques captured the streetscape factors from a pedestrian perspective and subsequently extracted various visual elements. In this study, independent variables include not only eight segmented object categories and two features detected from the streetscape images of community roads in Seoul but also factors relevant to the built environment and individual characteristics. In addition, human perceptions were measured through an online survey based on a 5-point Likert scale from 1 (“very dissatisfied”) to 5 (“very satisfied”) (n = 240). Subsequently, by using a multilevel ordered logistic regression model, we examined the factors’ discrete impacts on six different perceptual indicators: vitality, safety, beauty, boring, depression, and overall pedestrian satisfaction. The main results of this study are as follows. Among the segmented objects, sky, vegetation, wall, sidewalk, and pavement have important ramifications as perceptual indicators. In such a case, sky, vegetation, sidewalk, and pavement positively affect overall pedestrian satisfaction, vitality, safety, and beauty; however, the wall has an adverse impact on them. Moreover, the number of pedestrians and vehicles detected from the YOLOv3 algorithm is significantly associated with most perceptual indicators. Pedestrian volume is positively correlated with overall pedestrian satisfaction, vitality, and safety, whereas the effect of the number of vehicles is the opposite. Overall, though this study, we proposed policy implications to improve the walking environment.
URI
https://www.dbpia.co.kr/journal/articleDetail?nodeId=NODE10553252https://repository.hanyang.ac.kr/handle/20.500.11754/177168
ISSN
1226-7147;2383-9171
DOI
10.17208/jkpa.2021.04.56.2.79
Appears in Collections:
COLLEGE OF ENGINEERING[S](공과대학) > URBAN PLANNING AND ENGINEERING(도시공학과) > Articles
Files in This Item:
There are no files associated with this item.
Export
RIS (EndNote)
XLS (Excel)
XML


qrcode

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.

BROWSE