以Google街景探究都市環境之主客觀可步行性之關係

Abstract

城市規劃日益重視人行體驗，而步行環境的改善，已被證實能提高民眾的步行意願，並促進其身體健康。為了防止社會資源過分集中於特定區域而侵損到其他群體的利益，許多研究開始以可步行性的概念從環境正義的角度來審視城市內不同區域步行環境的資源分配狀況。過去研究多透過客觀環境或是主觀知覺衡量可都市環境之步行性，但這些研究多將客觀環境與主觀感知分開評估。雖有學者指出感知可步行性對行人步行體驗與步行意願等的反映較客觀可步行性更為準確，但感知可步行性也會受客觀環境的影響，卻鮮少研究探討可步行性之客觀環境與主觀知覺之關聯。而且，囿於現地調查成本較高，研究範圍侷限。拜科技進步所賜，街景服務普及，這些資訊能不僅提供街道立面環境屬性，且達全面性覆蓋，便於大範圍評估。 而最近的研究以語義分割計算街景各環境屬性面積，大幅減少了客觀環境評估成本，但鮮有研究真正探討到環境屬性對感知可步行性的具體影響程度。因此，本研究將運用Google街景研究都市環境之主客觀可步行性，同時利用語義分割與主觀感知建構環實質境屬性對感知可步行性之預測模型。並以此模型預測台北市各路段可步行性，關聯各地區的社經特性檢視台北市步行空間的環境正義問題。 本研究選擇臺北市作為研究案例，第一階段以中正區為抽樣地點，將區內道路按每50m為一取樣點，採分層隨機抽樣，共計取樣1,022點；再依道路四種功能分級，按25%比例進行抽樣，最後抽取256張街景圖片。在客觀可步行環境評估方面，運用由Cityscapes數據集訓練的深度卷積神經網絡模型(DeepLab v3+)，計算街景中人行道、樹等環境屬性面積佔比。另外，在主觀可步行性感知評估方面，則透過網路問卷進行使用者照片評估，請受測者針對照片內容以便利性、舒適度、安全性、吸引力四種指標評估感知可步行性。第二階段針對台北市非山坡地區域之道路按每50m為一取樣點進行普查，利用第一階段得到的預測模型進行感知可步行性得分預測，建立可步行性地圖，並以台北市各區以及各里別的收入、年齡、學歷之社經特性探討步行空間分佈的公平性。 第一階段的預測模型經由逐步回歸分析顯示，人行道、其他植物、喬木等對舒適感有顯著正面影響（R2=0.523）；人行道、立桿、道路對便利性有顯著正面影響，而牆壁對便利性則有顯著負面影響（R2=0.271）；人行道、其他植物、道路等對安全性有顯著正面影響，而牆壁、建築對便利性則有顯著負面影響（R2=0.513）；其他植物、道路、立桿對吸引力有顯著正面影響，而建築、卡車、牆壁對吸引力有顯著負面影響（R2=0.464）。將四種指標平均計算成平均感知而言，道路、其他植物、人行道等對感知可步行性有顯著正面影響，而牆壁則對感知可步行性有顯著負面影響（R2=0.501）。 第二階段可步行性地圖則顯示，台北市可步行性較高的街道主要位於主次要幹道及市地重劃區，且但台北市步行空間在里別級的分佈上存在一定的環境不正義現象，社經狀況較為弱勢的各里可步行性通常也較低。

Urban planning is increasingly focusing on the pedestrian experience, and improvements to the walking environment have been shown to increase people’s willingness to walk and improve their health. In order to prevent social resources from being overly concentrated in a specific area and infringing on the interests of other groups, many studies have begun to use the concept of walkability to examine the resource allocation of pedestrian environments in different areas of the city from the perspective of environmental justice. In the past, most studies measured the walkability of urban environments through objective environment or subjective perception, but most of these studies evaluated the objective environment and subjective perception separately. Although some scholars have pointed out that perceived walkability reflects pedestrians’ walking experience and willingness to walk more accurately than objective walkability, perceived walkability is also affected by the objective environment, but few studies have explored the relationship between the environment and subjective perception. Moreover, due to the high cost of on-site investigation, the research scope was limited. Thanks to the advancement of technology and the popularization of street view services, this information can not only provide the environmental attributes of street facades, but also achieve comprehensive coverage, which is convenient for large-scale evaluation. However, recent researches use semantic segmentation to calculate the area of each environmental attribute of streetscape, which greatly reduces the cost of objective environmental assessment, but few studies have really explored the specific impact of environmental attributes on perceived walkability. Therefore, this study will use Google Street View to study the subjective and objective walkability of urban environments, and at the same time use semantic segmentation and subjective perception to construct a predictive model of perceived walkability by the physical environment attributes of the environment. And using this model to predict the walkability of each road section in Taipei City, and examine the environmental justice issues of Taipei City's pedestrian space in relation to the social and economic characteristics of each region. This study selects Taipei City as a case study. In the first stage, Zhongzheng District is used as the sampling site, and every 50m of roads in the area is taken as a sampling point, and a stratified random sampling is used to sample a total of 1,022 points. Sampling is carried out at a ratio of 25%, and finally 256 street view images are selected. In terms of objective walkable environment assessment, the deep convolutional neural network model (DeepLab v3+) trained by the Cityscapes dataset is used to calculate the area ratio of environmental attributes such as sidewalks and trees in the street view. In addition, in terms of subjective walkability perception evaluation, the photos are evaluated through an online questionnaire, and the subjects are asked to evaluate the perceived walkability according to the content of the photo by four indicators: convenience, comfort, safety, and attractiveness. The second stage is to conduct a census on the roads in the non-hillside areas of Taipei City at every 50m sampling point, and use the prediction model obtained in the first stage to predict the perceived walkability score, build a walkability map, and use the districts in Taipei City. And the social and economic characteristics of income, age, and educational background of each distict and each village to explore the fairness of the distribution of walking space. The first-stage prediction model showed through stepwise regression analysis that sidewalks, ground cover, plants, etc. had a significant positive impact on comfort (R2=0.523); the number of destinations, sidewalks, poles, and roads had a significant positive impact on convenience. , and walls have a significant negative impact on convenience (R2=0.271); sidewalks, ground cover, roads, etc. have a significant positive impact on safety, while walls and buildings have a significant negative impact on convenience (R2=0.513) ; Ground cover, roads, and poles have a significant positive impact on attractiveness, while buildings, trucks, and fences have a significant negative impact on attractiveness (R2=0.464). Taking the average of the four indicators to calculate the perceived walkability, roads, ground cover plants, sidewalks, etc. have a significant positive impact on perceived walkability, while walls have a significant negative impact on perceived walkability (R2=0.501). The second-stage walkability map shows that the streets with high walkability in Taipei City are mainly located in the main and secondary arterial roads and urban rezoning areas. However, there is a certain degree of environmental injustice in the distribution of pedestrian spaces in Taipei City, and the walkability of each village with weaker socioeconomic status is usually lower.