EN中文

Journal of Geo-information Science >

2019 , Vol. 21 >Issue 1: 46 - 58

DOI: https://doi.org/10.12082/dqxxkx.2019.180311

A Review of Urban Environmental Assessment based on Street View Images

  • ZHANG Liying , 1, 2 ,
  • PEI Tao , 3, 4, * ,
  • CHEN Yijin 1 ,
  • SONG Ci 3 ,
  • LIU Xiaoqian 5
Expand
  • 1. College of Geoscience and Surveying Engineering, China University of Mining& Technology, Beijing 100083, China;
  • 2. College of Information Science and Engineering, China University of Petroleum, Beijing 102249, China
  • 3. State Key Laboratory of Resources and Environmental Information System, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
  • 4. University of Chinese Academy of Sciences, Beijing 100049, China
  • 5. College of Arts and Science of Beijing Union University, Beijing 100191, China
*Corresponding author: PEI Tao, E-mail:

Received date: 2018-07-04

  Request revised date: 2018-10-22

  Online published: 2019-01-20

Supported by

National Natural Science Foundation of China, No.41590845, 41525004, 41421001,41877523

Supported by a grant from State Key Laboratory of Resources and Environmental Information System

Copyright

《地球信息科学学报》编辑部 所有
Fold

Abstract

Urban environmental assessment research has traditionally adopted a method based on field survey, which is difficult to evaluate on a large scale and refined scale. Street view image has a wide coverage, can provide street-level landscape and intuitively reflect the city facade information, and have the advantage of lower cost than on-site data collection, so it provides a large sample data source and new research ideas for urban environmental assessment. Different from the sky view of remote sensing image and the user interaction data of geo-tagged social media, street view image is more focused on recording stereoscopic sectional view of the city street level from the perspective of people, which can represent scenes seen or felt from the ground on a fine scale, so it is more suitable to replace on-site observation of urban environmental assessment. The continuous breakthrough of artificial intelligence technology and its application in various fields make it possible to conduct urban environmental assessment research based on street view image on a wide range of spatial scales. In this paper, we first described and compared three categories of data sources commonly used in urban environmental assessment including street view image, remote sensing image and geo-tagged social media data, and summarized the advantages of street view image in urban environmental assessment. Then we classified the methods used in urban environment assessment based on street view image into the following four categories : methods based on image analysis, statistical analysis, artificial intelligence and spatial analysis. Next, from the urban physical, social, economic and aesthetic environment, we summarized the research and application of street view image in urban environmental assessment. Finally, we pointed out the innovations, limitations and future research directions of the urban environmental assessment based on street view image. On one hand, the application of artificial intelligence represented by deep learning promotes the research progress of urban environmental assessment on large-scale and fine-scale. On the other hand, in the era of big data, the integration of data source represented by street view image, remote sensing image, and geo-tagged social media data will help promote urban environmental assessment research from multiple perspectives and multi-level.

Key words： street view image; urban environmental assessment; urban factor; artificial intelligence; deep learning

Cite this article

ZHANG Liying , PEI Tao , CHEN Yijin , SONG Ci , LIU Xiaoqian . A Review of Urban Environmental Assessment based on Street View Images[J]. Journal of Geo-information Science, 2019 , 21(1) : 46 -58 . DOI: 10.12082/dqxxkx.2019.180311

1 引言

城市规模的扩大和城市化进程的加速,给城市发展带来巨大的压力。城市环境是指影响城市人类活动的各种自然的或人工的外部条件[1],是人类居住的基本场所和生活的基本条件,也是衡量城市发展重要的指标之一。狭义的城市环境是包括自然环境及人工环境组成的物理环境。广义的城市环境还包括社会环境、经济环境和美学环境。然而,随着城市化进程的加速和社会生产力的飞速发展,原有的自然环境状况在不同程度上被大量的人工环境所改变和取代,如道路、建筑、城市绿色景观等,城市的自然空间结构发生了巨大的变化,由完全的自然环境演变为结构复杂、功能齐全的人工化环境[2]。因此对城市环境进行客观地评价具有重大的意义,可以有效地调整城市职能、改善城市环境质量,协助制定环境保护的原则方案,确保城市经济与环境持续协调发展。
城市环境评价研究传统上采用基于现场调研的方法,难以在大范围、精细化的尺度上进行评估[3]。传感器技术和数字化技术的最新进展产生了新的数据采集手段,为城市环境评价研究提供高分辨率的实景图像大数据,如谷歌街景图像以高分辨率记录了大多数北美城市,这为大范围空间尺度上的城市物理特征的识别提供了新的机会[4]。相比于传统的现场调研数据而言,街景图像不但覆盖范围广、数据量大,而且地图商提供API开发接口,可以免费下载街景图像,更重要的是图像中包含的城市基础设施信息丰富,能够提供街道层级的人造景观和自然景观,直观准确地反应城市立面信息。这些优点使得街景图像成为城市环境评价研究中重要的新数据源[5,6],同时为城市环境评价研究带来了新的方法学机会。
基于街景图像的城市环境评价是根据环境评价的目标,利用街景图像提供的街道层级景观和立体轮廓信息,从中选择有代表性的评价要素,如道路、行人、树木和建筑物等,对街道、社区、城市等空间范围内的某一环境要素或综合环境要素进行分析评价[7],实现对城市物理环境、社会环境、经济环境或美学环境的评价。本文围绕基于街景图像进行的城市环境评价研究展开综述：首先对比了城市环境评价常用的3类数据源,归纳街景图像的特点;然后对城市环境评价过程中使用的方法进行分类;接着综述了街景图像在城市物理环境、社会环境、经济环境和美学环境评价中的应用研究进展;最后总结全文并对未来研究方向提出展望。

2 街景图像的特点

街景图像、遥感影像和地理标记社交媒体数据是目前城市环境评价中常用的3种数据源。街景全景已经在全球范围内广泛应用,研究中常用的街景图像数据源有：谷歌街景（Google Street View,GSV）[8]、百度街景（Baidu Street View,BSV）[9]和腾讯街景（Tencent Street View,TSV）[10]。GSV拥有全球网络,但不包括中国和其他一些国家,BSV和TSV把重点放在中国境内的全景拍摄上,已经覆盖了中国相当大的城市。通过从GSV,BSV和TSV 各自的公共API请求,可以免费获取静态街景图像。3种街景图像API的描述和对比见表1。
Tab. 1 Street view image APIs

表1 街景图像API

街景API 覆盖范围 图像最大分辨率（宽度×高度） 使用样例
谷歌 114个国家及地区 2048×2048 https://maps.googleapis.com/maps/api/streetview?size=400×400&location=40.720032,73.988354&fov=90&heading=235&pitch=10&key=YOUR_API_ KEY
百度 中国372座城市 1024×512 http://api.map.baidu.com/panorama/v2?width=512&height=256&location=116.313393,40.04778&fov=180&ak=YOUR_API_KEY
腾讯 中国296座城市 960×640 http://apis.map.qq.com/ws/streetview/v1/image?size=600×480&pano=10011 022120723095812200&pitch=0&heading=0&key= YOUR_API_KEY
城市街景图像具有如下优势：①街景图像以行人的视角详细系统地记录了城市街道级别的景象,图像中包含的城市基础设施信息丰富,提供了街道层级的人造景观和自然景观,能够直观准确地反映城市的立面信息;②街景图像覆盖范围广、数据量大。谷歌街景覆盖了114个国家和地区的城市,百度街景覆盖了中国372座城市,这为不同国家之间或者同一国家的不同城市之间的环境评价研究提供了坚实的数据源;③数据收集效率高,成本低。地图商提供API开发接口,可以免费下载街景图像,通过街景图像进行数据收集成本低,流程逻辑简单,同时便于监督和质量控制,并能解决一些现场采集数据的干扰性以及与实地调查高犯罪率街区的人员安全问题的担忧[11]。
遥感影像是通过遥感技术获取的高空地面影像,通过转换可以直观形象地展示地貌的形态,具有宏观、适时、快速获取地物的优点。城市环境具有非常强的时效性,如土地资源、植被覆盖变化、农作物分布、冰川变化、热岛环境等自然环境信息,使用常规的测量方法,耗时耗力,使用遥感影像可以省时省力地获取大范围内城市自然环境变化的图像资料和基本数据,如土壤、植被、大气、水体等信息,实现地表变化分析[12]、农作物识别[13]、空气质量评价[14]、灾情评价[15]和城市热环境变化[16]等城市自然环境的评价。
地理标记社交媒体数据是大数据时代地理信息的重要组成部分,包含空间、时间、语义等属性信息,具有数据量大、数据更新速度快,信息量丰富且与用户活动息息相关等优势。从中可以提取与环境相关的主题信息,结合地理位置、时间及发布的图片,可以实现空气质量[17]、台风灾害[18]、旅游景点评价[19]、城市风貌感知[20]等城市环境评价研究。
街景图像、遥感影像和地理标记社交媒体数据3种数据源的对比见表2。与遥感影像的天空俯视图和地理标记社交媒体的用户交互数据相比,街景图像更侧重于从人的视角记录城市街道层级的景象,它捕获到的是城市街景的立体剖面视图,且能以精细化的尺度记录从地面上人看到或感受到的景象,比如道路表面的平整度、清洁度等,建筑立面的视图、树木的垂直结构、街道的连续性、天空的开敞度等,可以说街景图像是城市风景的直接适宜表示。因此能代替现场观测的城市环境评价工作,尤其是对于在较大或地理上分散的地区进行的城市环境评价研究。
Tab. 2 Comparison of street view image, remote sensing image and geo-tagged social media data

表2 街景图像、遥感影像和地理标记社交媒体数据对比

数据类别 采样方式 优缺点 代表性研究
街景图像 地面拍摄 优点：从微观和人的视角精细化记录城市街道层级的立体剖面景象;覆盖范围广、数据量大、成本低
缺点：数据在空间分布不均匀		 社区环境[11]、城市安全感[21]、收入预测[22]、建筑特色[23]
遥感影像 空中拍摄 优点：从宏观和高空鸟瞰的视角记录城市,覆盖范围广、数据在空间分布均匀
缺点：成本高、分辨率低		 地表变化分析[12]、农作物识别[13]、空气质量评价[14]、灾情评价[15]、城市热环境变化[16]
地理标记社交
媒体数据		 网络用户发布 优点：用户交互的内容不仅有文本信息,还包含地理位置、时间、图像、视频、情感等信息;具有动态性、时效性和交互性;
缺点：数据稀疏、数据存在口语化、错误拼写和缩写、使用特殊符号等问题		 空气质量[17]、台风灾害[18]、旅游景点评价[19]、城市风貌感知[20]

3 街景图像在城市环境评价中使用的方法

城市环境评价是根据城市环境调查与监测资料,提取评价要素,应用各种评价方法对一个地区的环境质量做出的评定。按方法学的角度把环境评价过程中使用的方法,分为4类：基于图像分析的方法、基于统计分析的方法、基于人工智能的方法和基于空间分析的方法。

3.1 基于图像分析的方法

城市街景图像的要素提取主要以像元、对象、场景3种空间单元为研究对象,相应的要素提取方法分为3类：基于像元的图像分析方法、面向对象的图像分析方法和面向场景的图像分析方法。
（1）基于像元的图像分析方法。基于像元的图像分析是采用逐像元的方式,每个像元只能属于一个城市要素类别。郝新华等[3]采用基于像元的图像分析方法,将街景图片的色彩模式从RGB转换为HSV,然后从数字图像中提取各色相通道的值;针对每一个像元,计算像元的颜色在颜色光谱中的度数,从而实现绿视率的计算。Li等[24]在评估街道绿化时,采用基于像元的图像分析方法获得初始图像分类。
（2）面向对象的图像分析方法。与基于像元的方法相比,面向对象的方法处理单位是对象或图像分割,而不是单个像元。它先将图像分割为具有物理意义的同构多边形,然后根据每个多边形的光谱和几何特性将每个多边形分配到不同的类别[25]。因此,面向对象的分类方法有助于消除原始谷歌街景图像的光谱变化,并保持不同城市特征作为对象的完整性。因此,与基于像元的方法相比,面向对象的分类方法更适合于从GSV图像中提取绿色植被[25,26,27]。Li等[26]使用均值平移算法将GSV图像中的光谱R,G和B分量归一化到[0,1],进行图像分割,然后通过将每个对象的属性设置为3个RGB带中的该对象内的像元的平均值,生成新的专题图像,用于下一步提取绿色植被。Li等[28]基于对象的图像分析方法,从谷歌图像创建的鱼眼图像中提取天空区域。
（3）面向场景的图像分析方法。从数字照片中检测和量化城市环境场景的特征称之为场景理解,它可以用于客观量化真实世界中的特征和它们在景观中的空间分布以适应多种应用。基于像素和面向对象的图像分析提取的是地物的低层语义信息,而面向场景的图像分析提取的是地物的高层语义信息。场景通常用于识别城市功能区域或精细的城市土地利用模式等研究。Li[29]提出了一种基于场景分类算法和谷歌街景图像的块级别的详细土地利用信息推导方法。精度评估结果表明,该方法适用于区分住宅建筑和非住宅建筑。

3.2 基于统计分析的方法

统计学是基于数据分析来研究测定、收集、整理、归纳和分析反映数据的方法,并给出正确结论的科学。城市环境评价应用统计分析方法实现评价主题与街景图像要素之间的因果关系,常用的方法有如下3类：
（1）相关性分析法。相关性,也称为相关系数或关联系数,是用来衡量2个变量相对于其彼此独立的距离。在衡量街景图像能否代替现场环境审核的研究中,文献[11]使用Spearman排序相关性来验证基于街景图像审计和实际现场审核的一致性。文献[30]使用F方差分析和ICC组内相关系数来衡量步行和骑车的物理和虚拟审计之间的一致性。
（2）层次分析法。层次分析法是把一个复杂的问题按属性的逻辑关系逐层分解,形成一个层次结构来降低分析问题的难度,并在逐层分解的基础上加以综合,给出复杂问题的求解结果[31]。
（3）回归分析。回归分析是统计学分析数据的一种方法,是为了获得2个或多个变量间是否相关、相关方向以及相关强度,从而建立数学模型实现特定变量预测目标变量。回归分析是城市环境评价研究工作中常用的方法之一,主要用来研究环境评价主题和街景要素之间的关系。如使用多元线性回归分析研究感知安全与植被特征之间的关系[27];使用普通最小二乘法（OLS）多变量回归来模拟依赖（绿色视图指数）和自变量（社会变量）之间的关系[26]。

3.3 基于人工智能的方法

人工智能是研究开发用于模拟、延伸和扩展人的智能的理论、方法、技术及应用系统的一门新的技术科学[32]。机器学习是目前主流的人工智能实现方法。按学习方法划分,机器学习可以分为监督学习、无监督学习、半监督学习、深度学习、强化学习等。
（1）监督学习。监督学习是使用已有的训练样本训练获得一个最优模型,然后利用此模型将所有的输入映射为相应的输出,包括分类问题和回归问题。典型的算法有最近邻（KNN）、支持向量机（SVM）、决策树等。图像分析技术与机器学习相结合,成为大规模大范围内提取城市要素的有效方法[4]。Hazelhoff等[33]使用方向梯度直方图（HOG）和尺度不变特征变换（SIFT）相结合的方法提取街景图像中交通标志符号的特征集,然后使用SVM进行分类,实现了大规模检测街道图像的交通标志问题。Yin等[34]基于监督式学习思想,提出了聚合通道特征行人检测算法,实现自动从谷歌街景图像中提取信息,识别街道上的行人。
（2）无监督学习。事先没有任何训练样本,直接对数据进行建模。常见的应用场景包括关联规则的学习以及聚类等。常见算法包括Apriori算法以及k-Means算法。
（3）半监督学习。半监督学习是监督学习和无监督学习混合的结果,在这种“学习”中,算法需要一些训练数据,但是比监督学习少很多。
（4）深度学习。深度学习是一种机器学习形式,它使计算机能够从经验中学习,并根据概念层次理解世界[35]。深度学习除了可以学习特征和任务之间的关联以外,还能自动从简单特征中提取更加复杂的特征。随着人工智能和计算资源的不断发展,尤其是基于深度学习的计算机视觉技术在图像的识别与定位中取得了突破性的进展,准确率接近甚至在部分任务上已经超过了人类的水平。基于深度学习的图像语义分割可以对街景图像进行细致到区域与类别级别的理解,这为从街景图像中提取更为复杂的评价要素奠定了技术基础。Liu等[36]使用传统的SIFT直方图[37]和2个最先进的深度卷积网络,即Krizhevsky[38]和Szegedy[39]提取街景图像中的建筑物,实验结果表明两个深度卷积网络比传统的SIFT特性表现更好。Liu等[40]提出了一种用于从弱监督图像学习鲁棒视觉感知模型的分层深度多实例回归方法,该方法使用先进的神经网络作为基础模型,并引入了多实例回归网络来预测街景图像的安全分数。

3.4 基于空间分析的方法

空间分析是基于地理对象的位置和形态特征的空间数据分析技术[41]。空间分析可分为基于图的分析、基于数据的分析和基于事件机理的分析[42]。基于街景图像的城市环境评价,从街景图像中提取出城市环境要素,使用空间分析方法,研究城市环境要素的空间分布模式以及城市环境与城市要素之间的关系等。
（1）基于图的分析以图形操作为主,主要有叠加分析、缓冲区分析、网络分析等。郝新华等[3]基于腾讯街景对街道绿化进行量化,使用叠加分析和回归分析,探讨了绿视率与道路等级和区位的相关关系。
（2）基于数据的分析以空间统计学为理论基础。Salesses等[43]基于谷歌街景,使用Moran's I统计测量城市环境感知的空间分离;使用Getis空间滤波回归（GSFR）研究城市不平等和杀人之间的相互关系。Li等[44]基于谷歌街景,使用具有空间过滤的泊松回归模型来研究暴力犯罪与城市居住环境的物理特征之间的关系。
（3）基于事件机理的分析,直接使用由环境信息提供的初始边界条件的已有结果来描述事件,在此基础上根据研究主题进行深入分析。

4 街景图像在城市环境评价中的应用

使用街景图像和人工智能技术探索城市要素的量化和图像化表达,挖掘城市中额外的地理空间信息、以及更复杂或特定的指标,进而实现大范围、定量化的城市环境评价研究,已成为大数据时代城市环境评价研究的热点。广义的城市环境由物理环境、社会环境、经济环境、美学环境组成[1],而街景图像与直接拍摄到和人类知觉直接相关的场景信息以及显示与之相关的社会、经济、心理和身体上的感受有着一定联系。根据广义城市环境的组成,将街景图像在城市环境评价中的应用研究按城市物理环境评价、城市社会环境评价、城市经济环境评价和城市美景环境评价进行综述。使用街景图像开展的典型环境评价工作及所采用的评价要素、使用的数据集、方法和代表性研究见表3。
Tab. 3 Urban environment evaluation based on street view image

表3 基于街景图像的城市环境评价工作

类别 评价要素 数据集 方法 代表性研究
物理环境 绿色植物、行人安全、行人道设施、机动交通、建筑、交通标志等 谷歌街景 相关性分析
泊松回归
机器学习		 1.街景图像审计社区环境的可行性：Rundle[11],Badland[30],Clarke[45]; 2.城市安全性：Kronkvist[46],Li等[44],Mooney[47] 3.土地利用类型：Li[29]
社会环境 汽车、人行道、行人、建筑、天空等 谷歌街景 机器学习
深度学习		 1.人口分布与政治倾向：Gebru等[48];2.城市可步行性：Yin等[34],Hara等6],Yin[4];3.城市安全感：Porzi等[21],Li[27]
经济环境 绿色植被、地面,建筑物、树、天空 谷歌街景 基于像元的图像分析机器学习 1.收入预测：Glaeser[22];2. 收入与物理环境的关系：Li[26],Arietta[49]
美学环境 行道树、绿色植被、建筑物 谷歌街景百度街景
腾讯街景		 机器学习
深度学习
图像分析		 1.街道绿化：郝新华等[3],Berland[50],Li等[24]; 2. 城市风貌：Liu等[36],Cheng等[51] ,唐婧娴等[52]; 3. 建筑特色：Doersch等[53],Lee等[23]

4.1 城市物理环境评价

城市物理环境包括地形、地质、土壤、水文、气候、植物、动物等自然环境和房屋、道路、管线、基础设施、不同类型的土地利用、废气、废水等人工环境[1]。城市自然环境是构成城市环境的基础,而城市人工环境是实现城市各种功能所必需的物质基础设施。鉴于街景图像捕获到的城市要素更侧重于人工环境,如房屋、道路、基础设施、行道树等,因此,目前基于街景图像的城市物理环境评价研究主要是对人工环境的评价,主要包括：① 街景图像审计社区环境的可行性;② 城市安全性;③ 土地利用。
4.1.1 街景图像审计社区环境的可行性
社区环境是指在一定的地域范围社区所面对的及感受的自然环境条件和人工环境条件的总和。社区环境的质量与市民的健康[54]、城市的安全以及城市的可持续发展有着密切的关系,因此社区环境审计一直以来都是城市研究的主题之一。传统的研究方法是现场审核,耗时且昂贵[11],人力成本高且效率低,无法在大范围空间尺度上进行,而使用街景图像收集数据效率高,成本低,且覆盖范围广,可以高效地进行社区环境审计工作。
研究者首先对街景图像是否可以代替现场观测的环境审计的可行性进行了论证。Kelly等[7]使用谷歌街景图像构建评估所建环境特征代替实地收集数据评价环境,结果表明使用街景图像评价所建环境是一种可靠的方法。Badland等[30]对新西兰奥克兰4个街区的48条街道段进行审计,以审查与步行和骑车相关的建筑环境属性,采用现场调查和使用谷歌街景虚拟评估2种方法进行。研究发现一旦熟悉了谷歌街景软件,虚拟审计与实际的街道评估相比,效率更快;此外,大多数评估要素都表现出可以接受的一致程度。Wilson等[55]使用谷歌街景图像的分析结果表明,审核员能够可靠地解释环境清单中包含的至少84%的特征。Clarke等[45]研究发现虚拟审计工具可以提供娱乐设施,当地食物环境和土地一般使用的可靠指标。但是,使用谷歌街景评估更精细观察的要素（如垃圾或碎玻璃的存在）的可靠性较低。Rundle等[11]发现对于行人安全,机动交通和停车,以及活动基础设施等大多数项目具有很高的一致性。而随时间变化的要素,如人员、动物或垃圾等要素具有较低的一致性。另外,此项研究指出,使用街景图像无法评估静态视频图像本质上无法评估的现场审核要素,如噪音、气味和流量速度等。最近的研究采用计算机视觉技术,结合其它数据如人口和经济数据,对大范围空间尺度上的城市物理环境评价进行研究。Naik等[56]创建了一个囊括美国5个主要城市自然变化的高分辨率谷歌全景街景图像数据集,将城市外观的变化利用评分进行量化,并分析了影响社区外观提升的决定因素,最后用计算机视觉的方法检验目前几种城市外观变化的理论。
4.1.2 城市安全性
邻里环境紊乱水平一直被认为是邻里犯罪率和居民对犯罪恐惧程度的有力预测因素[57],包括建筑物的空间布局,街道设计和土地利用的多样性等物理特征[58]。Kronkvist[46]通过谷歌街景的虚拟系统社会观察,与自我报告的邻里环境紊乱水平相比,研究了在多大程度上可以可靠地审核邻里环境的物理紊乱。该研究由两组数据组成,通过瑞典马尔默市21个人口普查社区的谷歌街景的虚拟系统社会观察以及居住在特定社区的受访者的自我报告数据,对邻里环境物理紊乱的方法论多样性结构之间的相关性,以及实际观察到的物理紊乱与财产犯罪受害的关系进行了研究。结果表明,通过谷歌街景观察到的物理紊乱与自我报告的邻里物理紊乱水平及财产犯罪的受害有显著的相关性。Li等[44]使用谷歌街景研究了暴力犯罪与城市居住环境的物理特征之间的关系。用空间滤波和基于谷歌街景图像的环境审计的泊松回归方法解决虚拟环境审计目标地点的选择。通过应急表分析,确定了一系列物理环境因素。研究结果为犯罪和预防犯罪工作提供了理论和实践上的启示。
道路及行人环境改变,如改善照明,增加减速带,或保持路面标记可以显著提高行人的安全[59]。然而,这样的研究通常需要收集大量的数据,需要研究人员访问和编码研究中包含的每个十字交叉点。Mooney等[47]率先使用街景图像研究了这一主题,提出了一种基于信息技术的方法来评估与受伤相关的街道和十字路口的特征。该研究使用2007-2011年的谷歌街景图像,评估纽约市532个十字路口的9个特征与伤害频率之间的关系,结果表明通过信息技术方法发现的交通岛、视觉广告、公共汽车站和人行横道基础设施与纽约市的行人受伤人数增加有关,且研究结果与个人研究观察是一致的。该方法比基于位置的行人伤害研究花费的时间和费用更低,为虚拟实地考察行人伤害控制研究提供了一种可行且翔实的方法。
4.1.3 土地利用类型
土地利用类型图是城市规划管理的重要参考。然而,获得高分辨率城市用地地图是困难和费时的。Li等[29]基于机器学习和地理标记的街道级谷歌街景图像,提出了一种新的方法来获得土地利用信息。在纽约市的一个案例研究区,使用几种通用的图像特征来代表不同城市风景的街道级图像。在不同的街道图像特征的基础上,进一步利用机器学习对不同图像进行分类。精度评价结果表明,该方法是一种在建筑区块级别上进行土地利用制图的很有前景的方法。

4.2 城市社会环境评价

城市社会环境包括人口分布及动态、服务设施、娱乐设施、社会生活等,体现了城市区别于乡村及其它聚居形式的人类聚居区域,在满足人类在城市中各类活动方面所提供的条件[1]。通过识别街景图像中的城市要素,结合要素所映射的社会环境属性和意义,实现城市社会环境的评价研究。主要包括：① 人口分布与政治倾向;② 城市可步行性;③ 城市安全感。
4.2.1 人口分布与政治倾向
人口统计,对于一些人口大国,由于基数过于庞大,实际调研及最终成果展示中存在着很大的延迟。街景图像和深度学习相结合为此问题提供了高效的解决方法。基于谷歌街景,Gebru等[48]将人工智能的研究成果应用到人口统计学中。对谷歌街景图像中各社区的汽车类型和位置数据,使用基于深度学习的计算机视觉算法进行识别分析,研究成果为人口统计提供重大参考,其中还可以实现预测社区人员的政治倾向。基于谷歌街景的人工智能算法分析几乎可以实时生成分析结果,为美国全国人口统计需要巨额的花费和手工收集数据严重滞后的问题提供了有效地自动化实现方法。
4.2.2 城市可步行性
城市可步行性是一个复杂的结构,并没有一致的定义[60],从城市社会环境评价的角度而言,城市可步行性是衡量城市提供服务设施水平的一种方式,和市民的健康也有着密切的关系[61]。基于街景图像的可步行性评价研究侧重于提取有利于可步行性的城市环境特征,包括道路的平整、无障碍物、有助于诱导步行的周围环境及安全感等。人行道的可用性和质量显著地影响人们在城市环境中出行的方式[6]。传统上,人行道的质量评估是通过面对面的街道审核进行的,这是一项劳动密集型且成本高昂的工作。Hara等[6]探讨了使用众包方式标记谷歌图像的人行道可达性问题的可行性,并提出了一种有前景的新型高度可扩展的获取有关人行道无障碍的方法。建筑环境,作为街道的一个重要元素,会影响人们选择进行体育活动的行为[62]。Yin和Wan[4]探讨了如何将机器学习算法应用于谷歌街景图像,以帮助研究建筑环境的形态,特别是通过客观地生成一些与步行能力有关的街道级别城市设计特征,在3个方向上测量视觉封闭的天空比例,发现3种视觉封闭方法与行人流量和行走分数相关。行人数量是衡量行人量的一个量化指标,用于评估城市居民的步行能力以及与土地利用和其他建筑环境特征之间的相关性[63]。Yin等[34]提出基于计算机视觉的聚合通道特征行人检测技术,自动从谷歌街景图像中提取信息,识别街道上的行人,实现城市行人计数。该方法能够以较高的准确度快速地统计出行人数量,这为进行大规模的城市行人计数提供了新思路。
4.2.3 城市安全感
城市的视觉外观在塑造人类对周围城市环境的感知和反应方面起着核心作用。例如,城市空间的视觉质量影响其居民的心理状态,并可能导致负面的社会结果。因此,了解人们对城市空间的看法和评价变得极其重要。城市的安全感是人对城市场景感知的高级属性之一。Porzi等[21]提出了一种新的方法来预测谷歌街景图像中城市场景的安全感,是第一个引入CNN深度学习计算模型实现自动发现与城市感知相关的中级视觉模式,与以往方法的比较,显著提高了预测精度。Li等[27]以马萨诸塞州波士顿为研究区域,对绿色植被的可见度与安全感之间的关系进行了研究,发现在住宅用地、城市公共/机构用地、商业用地和露天用地中,提高绿色植被的可见度有助于提高居民在城市中的安全感。但是,同样的措施在运输和工业用地上可能并不奏效。对于不同的土地利用类型,选择适当类型的绿色植被可以使城市绿色植被提供的效益最大化,这对在空间非常有限的密集城市地区而言非常重要。

4.3 城市经济环境评价

城市经济环境包括资源、市场条件、就业、收入水平、经济基础、技术条件等,反映了城市经济发展的条件和潜能[1]。街景图像记录的是城市的物理环境,物理环境特征可以预测城市非视觉属性,进而结合城市空间物质属性的数据,如家庭收入数据,房价数据等,实现对预测的验证和经济环境的评价。主要包括：① 收入预测;② 收入与物理环境的关系。
4.3.1 收入预测
使用街景图像大数据源结合预测方法,对城市收入水平进行预测,实现对城市经济环境的评价。Glaeser等[22]使用计算机视觉模型从街景图像中预测纽约市居民收入的中位数。发现经过训练的计算机视觉模型能够预测纽约市的中位数收入,进而预测波士顿的中位数收入,其准确率几乎与纽约相同。研究还将预测收入与房价联系起来,显示该技术在房价回归中的潜在用途。
4.3.2 收入与物理环境的关系
物理环境特征和城市经济环境要素之间存在着关联关系。Li等[26]提出了一个修改后的绿色景观指数作为街道绿化的指标方法,研究根据不同水平和垂直角度捕获的谷歌图像,计算出绿色景观指数,以定量表示行人从地面可以看到多少绿色植物,进而探索城市街道绿化与美国康涅狄格州哈特福德的街区居民的地位（经济和种族/民族）之间的关系,发现不同社会条件的人在哈特福德的生活环境中有不同数量的街道绿化。收入较高的人往往生活在有更多街道绿化的地方。Arietta等[49]提出了一种自动识别和验证城市视觉外观与其非视觉属性（如犯罪统计数据,房价,人口密度等）之间预测关系的方法。给定一组街景图像和（位置,城市属性值）测量对,首先确定图像中区分属性的视觉元素。然后,通过使用非线性支持向量回归在这些元素上学习一组权重来训练一个预测器。对美国6个不同的城市的各种城市属性进行了预测。发现确实存在视觉元素与城市属性之间的预测关系,包括暴力犯罪率、盗窃率、房价、人口密度、树木存在、涂鸦存在以及对危险的认识。

4.4 城市美学环境评价

城市美学环境包括风景、风貌、建筑特色、文物古迹等,是城市形象、城市气质和韵味的外在表现和反映[1]。街景图像在大范围、精细化的尺度上记录了城市街道级别的立体剖面视图,可以细粒度地对城市美学环境进行评价。主要包括：① 街道绿化;② 城市风貌;③ 建筑特色。
4.4.1 街道绿化
城市街道绿地长期以来被认为是城市环境中重要的景观设计元素,而街景图像能以非常高的分辨率记录街道中不同层次高度的绿地植物,包括行道树、灌木、草坪,以及其他形式的植被,相比于测量城市绿化常用的遥感影像数据源,街景图像不会遗漏树冠下的灌木、草坪以及建筑物墙壁上的绿植。街道绿化是城市绿色空间的一个组成部分,在提高城市美景环境方面发挥着重要作用。绿视率是一个反映行人对周围绿色环境的感知程度指标,它是指绿化面积在行人正常视野面积中所占的比例[64]。城市居住区绿视率的计算方法通常以实地拍摄照片为数据源,然后计算照片中绿色区域在全图中所占比例[65]。但人工拍摄采样点有限,且不容易控制拍摄方向,难以开展大范围空间上的绿视率评价。而街景图像具有覆盖范围广、包含街道侧竖立面信息的优势,弥补了上述缺点。国外学者基于街景图像并结合图像分析技术,评估大规模范围内行人视角下对街道绿量的感知程度,不仅降低了人力物力,同时还提升了研究效率。Li等[24]对使用谷歌街景图像作为街道级城市绿化评估工具进行了探索,提出了一个修改现有的绿色景观指数（GVI）的公式,并使用谷歌街景图像在纽约市曼哈顿区东村地区对街道绿化进行了案例研究评估。研究发现谷歌街景非常适合评估街道绿化,修改后的GVI是一种相对客观的街道绿地测量方法。郝新华等[3]使用成都腾讯街景图像,提出一种自动实现大规模、精细化尺度的街道绿化的量化评价方法,研究结果表明成都街道以不绿和一般绿街道为主;街道绿化与道路等级、街道周边地块性质、区位等相关。
行道树是城市绿化的重要组成部分,它提供了宝贵的生态环境、社会和经济效益[66]。Berland等[50]使用谷歌街景对3个城市的行道树进行虚拟调查,并将结果与同一地点的现有实地数据进行比较。虚拟调查分析人员记录了街道树的位置,确定了树种到物种水平,并估计了胸径的直径。实地调查记录的597棵树中超过93%在虚拟调查中观察到。虚拟调查中的树识别在树种水平90%和物种水平66%上和实地数据一致。研究结果表明,街景视图中的虚拟调查可能适合生成某些类型的街道树数据,比现场调查能更有效地更新现有数据集。 Seiferling等[67]使用纽约和波士顿共计456 175个带有地理位置标记的谷歌街景图像,提出通过计算机视觉的方法来量化街道上的城市树木覆盖。该方法不仅从市民体验城市的角度展现了观察的城市景观,还在于从街景图像中获取的树高等垂直维度的信息,而这一信息恰恰是通过遥感影像（俯视角度）等传统手段不容易获得的。
4.4.2 城市风貌
城市风貌代表着城市的形象,可以反映出城市的特有景观和面貌、风采及神态。街景图像记录了城市的物理景观和面貌,可以从中提取反映城市风貌的要素,对城市风貌进行评价。Liu等[36]提出了一个包含3个机器学习模型,用于大规模和自动评估城市环境质量的方法。选取了北京五环内的百度街景图像对建筑立面的视觉质量和街墙的视觉连续性进行了大规模的研究,发现在视觉质量方面,城市规模明显,城市北上高于南部,特别是在城市边缘区域和五环路。北四环和北五环之间的大部分地区保持了现代化的城市形象,南部的许多地区更像是一个破旧的村庄,而不是一个城市。Cheng等[51]基于南京市的腾讯街景图像探讨了城市街道视觉感知分析的可能性,使用图像分割[68]和显著区域的检测[69]等图像处理方法,从大量街景图像中提取特征,进而分析并提出了4个能有效反映街道的视觉属性指标：显著区域饱和度、视觉熵绿色景观指数,和天空开敞度指数。实验结果表明,所提出的4项指标能有效地反映街道的视觉属性,可促进基于视觉感知的城市景观评估。唐婧娴等[52]基于大规模多时相、多角度街景图像数据,提出了“街道空间-品质评估-品质变化特征识别-影响因素分析”的框架,分析北京更新类居住区外围街道空间品质的水平和变化特征,探索大数据环境下街道空间品质评估的思路,来描述和评估城市街道空间品质。
4.4.3 建筑特色
建筑是城市景观的核心要素,建筑的历史地位、历史背景,以及它的地理位置和出色的艺术造型等,使其成为一个城市的象征。Doersch等[53]使用最近邻居技术,从谷歌街景提供的大型数据集中自动查找城市特征（窗户、灯等）,提出一种区分聚类方法来处理本地建筑身份的识别问题。Lee等[23]使用庞大的街道级图像集合来查找与特定时间段不同的语义级架构元素对应的视觉模式。使用这种分析方法来对建筑物进行日期分析,挖掘随着时间的推移功能相似的建筑元素（如窗户、门、阳台等）的风格如何发生变化。通过将来自巴黎的近 150 000幅Google街景图像的大型数据集与地籍图结合起来,验证巴黎建筑立面的大致施工日期,发现从1800年以前的非常平坦的结构开始演变到1851-1914年的多个装饰结构的越来越华丽的窗饰,19世纪50年代至40年代的长窗阳台演变到短阳台,1940年以后大型建筑物的栏杆,在1982-1989年转变为玻璃,在90年代转变为金属。

5 结论与讨论

街景图像更侧重于从人的视角记录城市街道层级的景象,数据覆盖范围广且采集数据成本低,是目前城市应用研究中重要的新数据源。本文关注基于街景图像进行的城市环境评价研究,分析对比了目前常用于城市环境评价的3种数据源,街景图像是城市风景的直接适宜表示,更适合于城市环境评价研究。研究结果发现与现场观测的街道评价相比,街景用于评价城市环境是一种效率更高的工具,且可实现大规模空间范围的城市环境评价和不同城市研究结果的比较。街景图像记录了城市的物理环境特征,可以直接反映城市物理环境和美学环境评价需要的要素,间接反映城市经济环境和社会环境评价的要素,因此对物理环境和美学环境评价的研究相对集中。最近几年,人工智能、计算资源的不断发展,机器学习与高性能计算的相结合,解决了短时间内成千上万张的街景图像的数据处理和信息提取问题,发展了基于人工智能技术的大范围空间尺度上的物理城市特征提取和图形化表达的新方法,促进了城市社会环境和经济环境评价的研究。
街景图像为城市环境评价研究带来新的数据支撑和研究思路的同时,也存在一些局限性。首先是时间变异性,由于街景环境的一些特征会随着时间发生变化,这些包括行人的数量、特征和活动, 以及停放或移动的车辆和许多诸如垃圾等物理紊乱的标志物。这些要素的情况和特征可能同时表现出随机变化和常规的日、季或天气有关的波动,测量误差可能会更高。此外,实地观察和街景拍摄的时间、季节、天气存在的不一致也可能导致偏差。例如,清晨收集街景视频可能会导致观察到的社会和行人活动水平降低,可能会（取决于垃圾清理的时间）影响测量街景的物理紊乱程度[11]。第二个局限性是时间一致性。街景图像并非都是在同一时期拍摄的,其中一些图像在冬季拍摄,其他图像在夏季或春季或秋季拍摄。在评估城市绿色空间的研究,需要同一时期拍摄的图像,为了保持时间一致性,需要检查图像,排除不是同一时期的数据。其次是隐私权的问题,为了保护街景图像中拍摄到的人的隐私权,街景图像中需要删除或模糊描绘人的属性的图像。这可能导致低估与邻里环境物理紊乱以及城市安全性研究有关的要素,带来研究结果的偏差。
鉴于城市街景图像的优势,街景图像已成为城市研究中新的数据源,使用街景图像研究城市环境评价也成为一种趋势。目前基于街景图像的城市环境评价研究还处于初期阶段,一方面以深度学习为代表的人工智能技术的发展,使得无法显式表达的模式识别变为可能;另一方面在大数据时代,以街景图像、遥感影像和地理标记社交媒体数据为代表的城市应用研究数据的融合,对城市环境评价研究发展新的理论和方法提出了挑战。未来基于街景图像的城市环境评价研究还应侧重如下3方面工作：
（1）融合各种数据源（如街景图像、遥感图像、地理标记社交媒体数据、手机信令和公交卡等）和新方法（深度学习、数据融合、大数据分析与可视化）,对城市环境进行多角度、多层次精细化尺度研究,提高城市环境评价的可靠性和精准性。
（2）基于街景图像的时间序列,对当前街景环境与历史环境的比较研究。不同时间节点的街景图像具有可比性,可以从历史街景图像中观察到环境特性,并且有助于研究时间动态变化的城市景观。除了专门使用街景图像之外,还可以携带移动设备并走进现场,使得基于街景的虚拟审计和现场观察能够同时进行,目前的城市环境可以直接与历史环境相比较[44]。
（3）基于街景图像的不同城市的环境评价对比研究。由于街景图像覆盖面广,百度、腾讯街景逐渐会覆盖全国更多城市,这为全国更大范围的城市环境评价提供了统一的数据来源。由于街景图像记录的城市物理环境特征侧重于人工环境,不包含诸如城市大气、土壤、水环境等自然环境特征,未来的研究需要重点解决如何将基于局部尺度街景图像的关键要素分析扩展到整个城市或全国多个城市等更大范围,并融合其他记录城市大气、土壤、水环境等自然环境要素的多元时空数据来集成分析城市环境,实现不同城市的环境评价。深度学习技术和计算资源的持续发展,也为研究全国各个城市的环境评价提供了技术支撑。这将为城市环境评价政策提供更加客观的参考数据。

The authors have declared that no competing interests exist.

References

Publishing order | Descend order by publishing year | Descend order by cited within
[1]
沈清基. 城市生态与城市环境[M].上海:同济大学出版社,1998.

[ Shen Q J.Urban ecology and urban environment[M]. Shanghai: Tongji University Press, 1998. ]

[2]
海热提·涂尔逊,杨志峰.试论城市环境与可持续发展[J].环境科学进展,1998,6(6):48-55.

[ Hai Re Ti·Tu E X, Yang Z F. Sustainable development advances in environmental science, 1998,6(6):48-55. ]

[3]
郝新华,龙瀛. 街道绿化:一个新的可步行性评价指标[J].上海城市规划,2017(1):32-36,49.街道绿化对居民尤其是行人的生活质量至关重要,是评估街道可步行性的重要指标之一。传统受限于数据获取的困难,对街道绿化的研究多局限于较小的地域,而目前运用新兴的街景图片进行评价的方法也多基于人工判断。通过构建一种自动的方法,实现了大规模、精细化尺度的街道绿化的量化评价,并以成都一、二圈层的街道为案例进行实践。研究发现:成都街道以不绿和一般绿街道为主;二圈层的街道绿化好于一圈层的街道;东部、北部个别街道绿化较好,南部、西部整体绿化相对较好;绿化相对较好的街道主要与大学、公园景点、河流两侧、居住区有关;街道绿化与道路等级、街道周边地块性质、区位等相关。对步行系统规划、街道品质改善提升等工作具有一定的指导意义。

[ Hao X H, Long Y.Street greenery: A new indicator for evaluating walkability[J]. Shanghai Urban Planning Review, 2017(1):32-36,49. ]

[4]
Yin L, Wang Z.Measuring visual enclosure for street walkability: Using machine learning algorithms and Google Street View imagery[J]. Applied Geography, 2016,76:147-153.

DOI

[5]
Dubey A, Naik N, Parikh D, et al.Deep learning the city: Quantifying urban perception at a global scale[C]. European Conference on Computer Vision, New York: Springer, 2016:196-212.

[6]
Hara K, Le V, Froehlich J.Combining crowdsourcing and google street view to identify street-level accessibility problems[C]. Proceedings of the SIGCHI conference on human factors in computing systems, ACM, 2013:631-640.

[7]
Kelly C M, Wilson J S, Baker E A, et al.Using google street view to audit the built environment: Inter-rater reliability results[J]. Annals of Behavioral Medicine, 2013,45(1):108-112.Observational field audits are recommended for public health research to collect data on built environment characteristics. A reliable, standardized alternative to field audits that uses publicly available information could provide the ability to efficiently compare results across different study sites and time.This study aimed to assess inter-rater reliability of built environment audits conducted using Google Street View imagery.In 2011, street segments from St. Louis and Indianapolis were geographically stratified to ensure representation of neighborhoods with different land use and socioeconomic characteristics in both cities. Inter-rater reliability was assessed using observed agreement and the prevalence-adjusted bias-adjusted kappa statistic (PABAK).The mean PABAK for all items was 0.84. Ninety-five percent of the items had substantial (PABAK a parts per thousand yenaEuro parts per thousand 0.60) or nearly perfect (PABAK a parts per thousand yenaEuro parts per thousand 0.80) agreement.Using Google Street View imagery to audit the built environment is a reliable method for assessing characteristics of the built environment.

DOI PMID

[8]
Google Street View (GSV)[EB/OL].

[9]
Baidu Street View (BSV)[EB/OL]. .

[10]
Tencent Street View (TSV)[EB/OL]. .

[11]
Rundle A G, Bader M D M, Richards C A, et al. Using google street view to audit neighborhood environments[J]. American Journal of Preventive Medicine, 2011,40(1):94-100.

DOI

[12]
Hansen M C, Loveland T R.A review of large area monitoring of land cover change using Landsat data[J]. Remote Sensing of Environment, 2012,122(1):66-74.Landsat data constitute the longest record of global-scale medium spatial resolution earth observation data. As a result, the current methods for large area monitoring of land cover change using medium spatial resolution imagery (10鈥50m) typically employ Landsat data. Most large area products quantify forest cover change. Forests are a comparatively easy cover type to map as well as a current focus of environmental monitoring concerning the global carbon cycle and biodiversity loss. Among existing change products, supervised or knowledge-based characterization methods predominate. Radiometric correction methods vary significantly, largely as a function of geographic/algorithmic scale. For instance, products created by mosaicking per scene characterizations do not require radiometric normalization. On the other hand, methods that employ a single index or classification model over an entire study area do require radiometric normalization. Temporal updating of cover change varies between existing products as a function of regional acquisition frequency, cloud cover and seasonality. With the Landsat archive opened for free access to terrain-corrected data, future product generation will be more data intensive. Per scene, interactive analyses will no longer be viable. Coupling free and open access to large data volumes with improved processing power will result in automated image pre-processing and land cover characterization methods. Such methods will need to leverage high-performance computing capabilities in advancing the land cover monitoring discipline. Robust validation efforts will be required to quantify product accuracies in determining the optimal change characterization methodologies.

DOI

[13]
宋盼盼,杜鑫,吴良才,等.基于光谱时间序列拟合的中国南方水稻遥感识别方法研究[J].地球信息科学学报,2017,19(1):117-124.多云多雾现象是农作物遥感分类经常遇到的问题,影响分类精度。为解决此类问题,本文提出一种基于时间序列GF-1号遥感影像识别水稻方法。利用多时相时间序列的GF-1号遥感影像提取中稻、晚稻的近红外波段(NIR)反射率、红光(R)波段反射率、归一化植被指数(NDVI)特征;拟合光谱和植被指数时间序列特征曲线;分析多时相影像离散近红外波段、红光波段、NDVI值落在拟合中稻、晚稻近红外波段、红光波段、NDVI时间序列曲线两侧的敏感性区域的比例,该区域也可以视为水稻作物识别特征的目标特征区域,只有达到一定的比例才能视为某类水稻作物。在此情形下,需要综合3种情况进行集中投票决定其最终分类结果。研究表明:该方法可以在多云雾地区对中稻和晚稻精确识别,中稻和晚稻用户精度可达95.97%和95.95%,总体精度为95.76%,kappa系数为0.9335。实验结果表明了NIR、R、NDVI时间序列曲线拟合的有效性,以及拟合曲线目标特征区域设置的合理性。

DOI

[ Song P P, Du X, Wu L C, et al.Research on the method of rice remote sensing identification based on spectral time-series fitting in southern China[J]. Journal of Geo-information Science, 2017,19(1):117-124. ]

[14]
Olaguer E P, Stutz J, Erickson M H, et al.Real time measurement of transient event emissions of air toxics by tomographic remote sensing in tandem with mobile monitoring[J]. Atmospheric Environment, 2017,150:220-228.During the Benzene and other Toxics Exposure (BEE-TEX) study, a remote sensing network based on long path Differential Optical Absorption Spectroscopy (DOAS) was set up in the Manchester neighborhood beside the Ship Channel of Houston, Texas in order to perform Computer Aided Tomography (CAT) scans of hazardous air pollutants. On 18-19 February 2015, the CAT scan network detected large nocturnal plumes of toluene and xylenes most likely associated with railcar loading and unloading operations at Ship Channel petrochemical facilities. The presence of such plumes during railcar operations was confirmed by a mobile laboratory equipped with a Proton Transfer Reaction-Mass Spectrometer (PTR-MS), which measured transient peaks of toluene and C-benzenes of 50 ppb and 57 ppb respectively around 4 a.m. LST on 19 February 2015. Plume reconstruction and source attribution were performed using the 4D variational data assimilation technique and a 3D micro-scale forward and adjoint air quality model based on both tomographic and PTR-MS data. Inverse model estimates of fugitive emissions associated with railcar transfer emissions ranged from 2.0 to 8.2 kg/hr for toluene and from 2.2 to 3.5 kg/hr for xylenes in the early morning of 19 February 2015.

DOI

[15]
苏亚丽,郭旭东,雷莉萍,等.基于多源卫星遥感的暴雨灾情时空动态信息的提取[J].地球信息科学学报,2018,20(7):1004-1013.

[ Su Y L, Guo X D, Lei L P, et al.Spatio-temporal dynamics of the impacts of rainstorm disaster on crop growing using multi-satellites remote sensing[J]. Journal of Geo-information Science, 2018,20(7):1004-1013. ]

[16]
侯浩然,丁凤,黎勤生.近20年来福州城市热环境变化遥感分析[J].地球信息科学学报, 2018,20(3):385-395.城市热环境是城市微气候的重要组成部分,已成为近年来的研究热点。受制于卫星传感器较低的热红外波段空间分辨率,此类数据反演得到的地表温度难以反映城市热环境的实际情况。为解决这一困境,本文利用空间降尺度HUTS算法反演得到30 m空间分辨率的福州市中心城区1994年5月12日、2003年5月29日和2016年7月27日3个时相的地表温度影像。在此基础上,结合土地利用等数据对热环境的时空变化做定量分析,并进一步引入景观指数,分析近20年间福州市中心城区高温度等级斑块的形态变化。结果表明:(1)近20年间随着城市拓展,福州市建成区的高温区域面积从35.75 km2增加到184.11 km2,高温度等级斑块不断从市中心向四周扩散;(2)市中心的特高温斑块和高温斑块趋向破裂、分散,聚集程度下降,次高温斑块的面积与占比均大幅提升,成为建成区内高温区域的主要组成部分;(3)城市热岛比例指数URI由0.39上升到0.52,热岛效应明显加强。总体上,近20 a间福州市建成区的热环境变化较大,其中鼓楼区南部、台江区和晋安区南部的高温区域聚集现象有所改善,而仓山区、马尾区和闽侯县的大部分区域在经历快速城市化过程后温度等级明显升高。

[ Hou H R, Ding F, Li Q S. Remote sensing analysis of changes of urban thermal environment of Fuzhou City in China in the past 20 years[J]. Journal of Geo-information Science, 2018,20(3):385-395. ]

[17]
Wang S, Paul M J, Dredze M.Social media as a sensor of air quality and public response in China[J]. Journal of medical Internet research, 2015,17(3):e22.Recent studies have demonstrated the utility of social media data sources for a wide range of public health goals, including disease surveillance, mental health trends, and health perceptions and sentiment. Most such research has focused on English-language social media for the task of disease surveillance. We investigated the value of Chinese social media for monitoring air quality trends and related public perceptions and response. The goal was to determine if this data is suitable for learning actionable information about pollution levels and public response. We mined a collection of 93 million messages from Sina Weibo, China largest microblogging service. We experimented with different filters to identify messages relevant to air quality, based on keyword matching and topic modeling. We evaluated the reliability of the data filters by comparing message volume per city to air particle pollution rates obtained from the Chinese government for 74 cities. Additionally, we performed a qualitative study of the content of pollution-related messages by coding a sample of 170 messages for relevance to air quality, and whether the message included details such as a reactive behavior or a health concern. The volume of pollution-related messages is highly correlated with particle pollution levels, with Pearson correlation values up to .718 (n=74, P<.001). Our qualitative results found that 67.1% (114/170) of messages were relevant to air quality and of those, 78.9% (90/114) were a firsthand report. Of firsthand reports, 28% (32/90) indicated a reactive behavior and 19% (17/90) expressed a health concern. Additionally, 3 messages of 170 requested that action be taken to improve quality. We have found quantitatively that message volume in Sina Weibo is indicative of true particle pollution levels, and we have found qualitatively that messages contain rich details including perceptions, behaviors, and self-reported health effects. Social media data can augment existing air pollution surveillance data, especially perception and health-related data that traditionally requires expensive surveys or interviews.

DOI PMID

[18]
梁春阳,林广发,张明锋,等.社交媒体数据对反映台风灾害时空分布的有效性研究[J].地球信息科学学报,2018,20(6):807-816.当灾害事件发生时,与之相关的社交媒体数据不断产生,其中包含了丰富的灾情信息和签到地理位置信息,这为灾情态势的及时感知提供了一种新的数据源,但是因社交媒体用户量的地区差异及网络空间中信息传播模式的特点,给社交媒体签到数据所代表的空间点过程的模式分析带来了一些新的问题,如签到点密度与实际灾害点事件密度之间的对应关系、签到点之间的空间关系、点格局的空间异质性及其影响因素等。本文以2016年14号台风“莫兰蒂”为例,以“台风”和“莫兰蒂”为关键词,在新浪微博平台上采集了2016年9月14-17日的微博数据,使用文档主题生成模型（Latent Dirichlet Allocation, LDA）和支持向量机（Support Vector Machine, SVM）对微博文本进行分类,构建了含有签到位置信息的灾情点事件数据库。在此基础上,针对社交媒体用户分布的空间异质性提出了一种基于签到点用户活跃度的加权模型。以全局自相关统计量Moran′s 为指标,对加权前后的签到微博数据进行对比,发现这些在社交网络中产生的签到微博数据在现实地理空间中存在明显的空间自相关性;基于“雨”、“停电”等关键词,利用上述加权处理后的微博数据库进行灾害制图,结合真实灾情资料进行时空对比分析,结果表明系列图谱能够反映台风灾害的时空过程趋势。

[ Liang C Y, Lin G F, Zhang M F, et al.Assessing the effectiveness of social media data in mapping the distribution of Typhoon disasters[J]. Journal of Geo-information Science, 2018,20(6):807-816. ]

[19]
刘逸,保继刚,朱毅玲.基于大数据的旅游目的地情感评价方法探究[J].地理研究,2017,36(6):1091-1105.基于情绪分类取向,通过界定三个旅游文本情感分析的过滤参数:旅游专属词库、语义逻辑规则和情感乘数,构建基于网络大数据的旅游目的地情感评价模型.基于该模型,抓取了120731条游客评论对8个旅游目的地进行评价,并以联合国世界旅游组织旅游可持续发展监测数据作为标准数据进行校验.研究证实三个过滤参数具有一定的科学性,能够较为准确地捕捉到游客对目的地评价的总体情感意象;经过单年度和多年度校验,六类规则的准确度依次为:C2＞C1 ＞C3＞B＞评分法＞A,即规则C2下的评价结果与监测结果最为吻合.结论证实了旅游大数据的可用性,为后续的理论推进和实践应用提供了科学依据.

DOI

[ Liu Y, Bao J G, Zhu Y L.Exploring emotion methods of tourism destination evaluation: A big-data approach[J]. Geographical Research, 2017,36(6):1091-1105. ]

[20]
易峥,李继珍,冷炳荣,等.基于微博语义分析的重庆主城区风貌感知评价[J].地理科学进展,2017,36(9):1058-1066.采用社交网络数据对城市进行感知与评价是一种新的城市感知定量研究方法。本文将新浪微博签到数据作为城市感知的数据来源,通过文本挖掘和语义分析方法,探索用户对重庆主城区城市风貌的感知与评价。针对重庆主城区特色风貌片区,研究形成了签到地图、情绪地图和对象地图。签到地图客观反映重庆主城区新浪微博用户活跃程度的空间分布特征;情绪地图挖掘活跃用户对空间的情绪表达和正负面态度;对象地图解析出现这种签到热力分布和情绪分布的原因,发现引发正负面态度的情绪对象。通过分析用户对规划师所设定的风貌要素载体的典型意见,将要素载体归纳为有感知积极、有感知消极、无感知和有感知未设定4类。今后应进一步从要素识别和价值判断两个方面为规划师塑造和管控城市风貌提供参考意见。

DOI

[ Yi Z, Li J Z, Leng B R, et al.Perception and evaluation of cityscape characteristics using semantic analysis on microblog in the main urban area of Chongqing municipality[J]. Progress In Geography, 2017,36(9):1058-1066. ]

[21]
Porzi L, Rota Bulò S, Lepri B, et al.Predicting and understanding urban perception with convolutional neural networks[C]. Proceedings of the 23rd ACM international conference on Multimedia, ACM, 2015:139-148.

[22]
Glaeser E L, Kominers S D, Luca M, et al.Big data and big cities: The promises and limitations of improved measures of urban life[J]. Economic Inquiry, 2018,56(1):114-137.New, “big data” sources allow measurement of city characteristics and outcome variables at higher collection frequencies and more granular geographic scales than ever before. However, big data will not solve large urban social science questions on its own. Big urban data has the most value for the study of cities when it allows measurement of the previously opaque, or when it can be coupled with exogenous shocks to people or place. We describe a number of new urban data sources and illustrate how they can be used to improve the study and function of cities. We first show how Google Street View images can be used to predict income in New York City, suggesting that similar imagery data can be used to map wealth and poverty in previously unmeasured areas of the developing world. We then discuss how survey techniques can be improved to better measure willingness to pay for urban amenities. Finally, we explain how Internet data is being used to improve the quality of city services. ( JEL R1, C8, C18)

DOI

[23]
Lee S, Maisonneuve N, Crandall D, et al.Linking past to present: Discovering style in two centuries of architecture[C]. IEEE International Conference on Computational Photography, IEEE, 2015:1-10.

[24]
Li X, Zhang C, Li W, et al.Assessing street-level urban greenery using Google Street View and a modified green view index[J]. Urban Forestry & Urban Greening, 2015,14(3):675-685.Highlights 61 Google Street View was used for assessment of street-level urban greenery. 61 Modified green view index was proposed and compared with vegetation coverage to assess the spatial distribution street-level urban greenery. 61 Google Street View proved to have great potential for future urban environmental planning. Abstract We explored Google Street View (GSV) as a street-level, urban greenery assessment tool. Street-level greenery has long played a critical role in the visual quality of urban landscapes. This living landscape element can and should be assessed for the quality of visual impact with the GSV information, and the assessed street-level greenery information could be incorporated into urban landscape planning and management. Information on street-level views of urban greenery assessment, however, is rare or nonexistent. Planners and managers’ ability to plan and manage urban landscapes effectively and efficiently is, therefore, limited. GSV is one tool that might provide street-level, profile views of urban landscape and greenery, yet no research on GSV for urban planning seems available in literature. We modified an existing Green View Index (GVI) formula and conducted a case study assessment of street greenery using GSV images in the area of East Village, Manhattan District, New York City. We found that GSV to be well suited for assessing street-level greenery. We suggest further that the modified GVI may be a relatively objective measurement of street-level greenery, and that GSV in combination with GVI may be well suited in guiding urban landscape planning and management. Graphical abstract The calculation of modified green view index and the discrepancy of spatial distribution of modified green view index and vegetation cover map. The sizes of the solid dots represent the magnitudes of green view index values. Figure options Download full-size image Download as PowerPoint slide

DOI

[25]
Li X, Meng Q, Gu X, et al.A hybrid method combining pixel-based and object-oriented methods and its application in Hungary using Chinese HJ-1 satellite images[J]. International Journal of Remote Sensing, 2013,34(13):4655-4668.Pixel-based and object-oriented processing of Chinese HJ-1-A satellite imagery (resolution 30 m) acquired on 23 July 2009 were utilized for classification of a study area in Budapest, Hungary. The pixel-based method (maximum likelihood classifier for pixel-level method (MLCPL)) and two object-oriented methods (maximum likelihood classifier for object-level method (MLCOL) and a hybrid method combining image segmentation with the use of a maximum likelihood classifier at the pixel level (MLCPL)) were compared. An extension of the watershed segmentation method was used in this article. After experimenting, we chose an optimum segmentation scale. Classification results showed that the hybrid method outperformed MLCOL, with an overall accuracy of 90.53%, compared with the overall accuracy of 77.53% for MLCOL. Jeffries atusita distance analysis revealed that the hybrid method could maintain spectral separability between different classes, which explained the high classification accuracy in mixed-cover types compared with MLCOL. The classification result of the hybrid model is preferred over MLCPL in geographical or landscape ecological research for its accordance with patches in landscape ecology, and for continuity of results. The hybrid of image segmentation and pixel-based classification provides a new way to classify land-cover types, especially mixed land-cover types, using medium-resolution images on a regional, national, or global basis.

DOI

[26]
Li X, Zhang C, Li W, et al.Who lives in greener neighborhoods? The distribution of street greenery and its association with residents' socioeconomic conditions in Hartford, Connecticut, USA[J]. Urban Forestry & Urban Greening, 2015,14(4):751-759.Street greenery plays an important role in enhancing the environmental quality of a city. Current urban environmental studies mainly focus on the distribution of desirable land uses (e.g., open spaces and parks). Few studies have been conducted on street greenery in residential areas, although it may provide a series of benefits to urban residents, such as energy saving, provision of shade, and aesthetic values. Google Street View (GSV) provides profile views of urban landscapes, and thus may be used for residential street greenery assessment. In this project, GSV was used in a case study to examine the relationships between the spatial distributions of residential street greenery and some socioeconomic variables in different block groups of Hartford, Connecticut, USA. The green view index was calculated based on the GSV images captured at different horizontal and vertical view angles to quantitatively represent how much greenery a pedestrian can see from ground level. Results showed that people with various social conditions have different amounts of street greenery in their living environments in Hartford. People with higher incomes tend to live in places with more street greenery. In summary, this study makes contribution to literature by providing insights into the living environments of urban residents in terms of street greenery, and it also generates valuable reference data for future urban greening programs.

DOI

[27]
Li X, Zhang C, Li W.Does the visibility of greenery increase perceived safety in urban areas? Evidence from the place pulse 1.0 dataset[J]. ISPRS International Journal of Geo-information, 2015,4(3):1166-1183.Land-use maps are important references for urban planning and urban studies. Given the heterogeneity of urban land-use types, it is difficult to differentiate different land-use types based on overhead remotely sensed data. Google Street View (GSV) images, which capture the fa ades of building blocks along streets, could be better used to judge the land-use types of different building blocks... [Show full abstract]

DOI

[28]
Li X, Ratti C, Seiferling I.Quantifying the shade provision of street trees in urban landscape: A case study in Boston, USA, using Google Street View[J]. Landscape & Planning, 2018,169:81-91.

[29]
Li X, Zhang C, Li W.Building block level urban land-use information retrieval based on Google street view images[J]. Giscience & Remote Sensing, 2017,54(6):819-835.Abstract Land-use maps are important references for urban planning and urban studies. Given the heterogeneity of urban land-use types, it is difficult to differentiate different land-use types based on overhead remotely sensed data. Google Street View (GSV) images, which capture the fa ades of building blocks along streets, could be better used to judge the land-use types of different building blocks based on their fa ade appearances. Recently developed scene classification algorithms in computer vision community make it possible to categorize different photos semantically based on various image feature descriptors and machine-learning algorithms. Therefore, in this study, we proposed a method to derive detailed land-use information at building block level based on scene classification algorithms and GSV images. Three image feature descriptors (i.e., scale-invariant feature transform-Fisher, histogram of oriented gradients, GIST) were used to represent GSV images of different buildings. Existing land-use maps were used to create training datasets to train support vector machine (SVM) classifiers for categorizing GSV images. The trained SVM classifiers were then applied to case study areas in New York City, Boston, and Houston, to predict the land-use information at building block level. Accuracy assessment results show that the proposed method is suitable for differentiating residential buildings and nonresidential buildings with an accuracy of 85% or so. Since the GSV images are publicly accessible, this proposed method would provide a new way for building block level land-use mapping in future.

DOI

[30]
Badland H M, Opit S, Witten K, et al.Can virtual streetscape audits reliably replace physical streetscape audits?[J]. Journal of Urban Health, 2010,87(6):1007-1016.

DOI

[31]
谢家平. 绿色设计评价与优化[M].北京:中国地质大学出版社,2004.

[ Xie J P.Green design evaluation and optimization[M]. Beijing: China University of Geosciences Press, 2004. ]

[32]
苗夺谦,张清华,钱宇华,等.从人类智能到机器实现模型——粒计算理论与方法[J].智能系统学报,2016,11(6):743-757.人工智能是研究、开发用于模拟、延伸和扩展人的智能的理论、方法、技术及应用系统的一门新的技术科学,是对人的意识、思维过程的模拟.粒计算是当前智能信息处理领域中一种新的概念和计算范式,是研究基于多层次粒结构的思维方式、复杂问题求解、信息处理模式及其相关理论、技术和工具的方法论.本文首先分析了人工智能模拟人脑智能的粒计算模式与方法,其次总结了粗糙集、商空间、模糊集、云模型、三支决策等几种典型的粒计算基本构架与数学模型,然后分析知识的多粒度解析表示与不确定性度量的研究现状,最后展望了粒计算求解模式在大数据时代所面临的机遇与挑战.

DOI

[ Miao D Q, Zhang Q H,Qian Y H,et al.From human intelligence to machine implementation model: Theories and applications based on granular computing[J].CAAI Transactions on Intelligent Systems,2016,11(6):743-757. ]

[33]
Hazelhoff L, Creusen I M.Exploiting street-level panoramic images for large-scale automated surveying of traffic signs[J]. Machine vision and applications, 2014,25(7):1893-1911.Accurate and up-to-date inventories of traffic signs contribute to efficient road maintenance and a high road safety. This paper describes a system for the automated surveying of road signs from street-level images. This is an extremely challenging task, as the involved capturings are non-densely sampled, captured under a wide range of weather conditions and signs may be distorted. The described system is designed in a generic and learning-based fashion, which enables the recognition of different sign appearance classes with the same algorithms, based on class-specific training data. The system starts with detection of the signs visible within each image, using a detection cascade. Next, the 3D position of the signs that are detected consequently within consecutive capturings is calculated. Afterwards, each positioned road sign is classified to retrieve its sign type, thereby exploiting all detections used during positioning of the respective sign. The presented system is intended for large-scale application and currently supports 11 sign appearance classes, containing 176 different sign types. Performance evaluations conducted on a large, real-world dataset (68,010 images) show that our approach accurately positions 95.5 % of the 3,385 present signs, where 96.3 % of them are also correctly classified. Furthermore, our system localized 98.5 % of the signs in at least a single image. Our system design allows for appending a limited manual correction stage to attain a very high performance, so that sign inventories can be created cost effectively.

DOI

[34]
Yin L, Cheng Q, Wang Z, et al."Big data" for pedestrian volume: Exploring the use of Google Street View images for pedestrian counts[J]. Applied Geography, 2015,63:337-345.61Google Street View provides street images in seven different resolution levels.61Level 3 images should be used for automatic pedestrian detection, if the default ACF training set and parameters are used.61A re-training is needed if level 4 or higher images are to be used to match the parameters of ACF and the Google images.61The image detection method proposed is capable of determining the presence of pedestrian with a reasonable accuracy level.61This method can help to get an objective and large scale reliable estimate of pedestrian volume.

DOI

[35]
Goodfellow I, Bengio Y, Courville A, et al.Deep learning[M]. Cambridge: MIT press Cambridge, 2016.

[36]
Liu L, Silva E A, Wu C, et al.A machine learning-based method for the large-scale evaluation of the qualities of the urban environment[J]. Computers, Environment and Urban Systems, 2017,65:113-125.Given the present size of modern cities, it is beyond the perceptual capacity of most people to develop a good knowledge about the qualities of the urban space at every street corner. Correspondingly, for planners, it is also difficult to accurately answer questions such as ‘where the quality of the physical environment is the most dilapidated in the city that regeneration should be given first consideration’ and ‘in fast urbanising cities, how is the city appearance changing’. To address this issue, in the present study, we present a computer vision method that contains three machine learning models for the large-scale and automatic evaluation on the qualities of the urban environment by leveraging state-of-the-art machine learning techniques and wide-coverage street view images. From various physical qualities that have been identified by previous research to be important for the urban visual experience, we choose two key qualities, the construction and maintenance quality of building facade and the continuity of street wall, to be measured in this research. To test the validity of the proposed method, we compare the machine scores with public rating scores collected on-site from 752 passers-by at 56 locations in the city. We show that the machine learning models can produce a medium-to-good estimation of people's real experience, and the modelling results can be applied in many ways by researchers, planners and local residents.

DOI

[37]
Lowe D G.Object recognition from local scale-invariant features[C]. The proceedings of the seventh IEEE international conference on, IEEE, 1999:1150-1157.

[38]
Krizhevsky A, Sutskever I, Hinton G E.Imagenet classification with deep convolutional neural networks[C]. Advances in neural information processing systems, ACM,2012:1097-1105.

[39]
Szegedy C, Liu W, Jia Y, et al.Going deeper with convolutions[C]. Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, IEEE, 2015:1-9.

[40]
Liu X, Chen Q, Zhu L, et al.Place-centric Visual Urban Perception with Deep Multi-instance Regression[C]. Proceedings of the 2017 ACM on multimedia conference, ACM, 2017:19-27.

[41]
郭仁忠. 空间分析[M].北京:高等教育出版社,2001.

[ Guo R Z. Spacial analysis[M]. Beijing: Higher Education Press, 2001. ]

[42]
杨志恒. GIS空间分析研究进展综述[J].安徽农业科学,2012,40(3):1918-1919.主要分析GIS空间分析的内容、技术方法、应用模型和实现方式,并对未来发展进行了展望。

DOI

[ Yang Z H.Review on research progress of GIS spatial analysis[J]. Journal of Anhui Agricultural Sciences, 2012,40(3):1918-1919. ]

[43]
Salesses P, Schechtner K, Hidalgo C A.The collaborative image of the city: Mapping the inequality of urban perception[J]. PloS one, 2013,8(7):e68400.A traveler visiting Rio, Manila or Caracas does not need a report to learn that these cities are unequal; she can see it directly from the taxicab window. This is because in most cities inequality is conspicuous, but also, because cities express different forms of inequality that are evident to casual observers. Cities are highly heterogeneous and often unequal with respect to the income of their residents, but also with respect to the cleanliness of their neighborhoods, the beauty of their architecture, and the liveliness of their streets, among many other evaluative dimensions. Until now, however, our ability to understand the effect of a city's built environment on social and economic outcomes has been limited by the lack of quantitative data on urban perception. Here, we build on the intuition that inequality is partly conspicuous to create quantitative measure of a city's contrasts. Using thousands of geo-tagged images, we measure the perception of safety, class and uniqueness; in the cities of Boston and New York in the United States, and Linz and Salzburg in Austria, finding that the range of perceptions elicited by the images of New York and Boston is larger than the range of perceptions elicited by images from Linz and Salzburg. We interpret this as evidence that the cityscapes of Boston and New York are more contrasting, or unequal, than those of Linz and Salzburg. Finally, we validate our measures by exploring the connection between them and homicides, finding a significant correlation between the perceptions of safety and class and the number of homicides in a NYC zip code, after controlling for the effects of income, population, area and age. Our results show that online images can be used to create reproducible quantitative measures of urban perception and characterize the inequality of different cities.

DOI PMID

[44]
Li H, Páez A, Liu D.Built environment and violent crime: An environmental audit approach using Google Street View[J]. Computers Environment & Urban Systems, 2017,66:83-95.漏 2017 Elsevier Ltd Recent studies empirically support the role of the built environment in inducing or hindering violent crime. Particularly, studies of the broken window theory have provided evidence that physical disorder is an environmental correlate of crime. This includes broken windows, vacant/abandoned housings, abandoned cars on street, graffiti, and decayed street lighting, among other things. Current studies are limited by the difficulty involved in collecting fine-scale quantitative environmental data. The conventional environmental audit approach, which aims to assess environmental features, is costly, time-consuming, and burdensome. In this study, we use Google Street View to study the relationship between violent crime and physical features of urban residential environment. More concretely, a Poisson regression model with spatial filtering is used to identify socio-economic correlates of violent crime. Parting from the hypothesis that omission of built environmental factors results in systematic residual pattern, we proceed to analyze the spatial filter to select sites for virtual environmental audits. A series of physical environmental factors are identified using contingency table analysis. The results provide both theoretical and practical implications for several theories of crime and crime prevention efforts.

DOI

[45]
Clarke P, Ailshire J, Melendez R, et al.Using Google Earth to conduct a neighborhood audit: Reliability of a virtual audit instrument[J]. Health & Place, 2010,16(6):1224-1229.Over the last two decades, the impact of community characteristics on the physical and mental health of residents has emerged as an important frontier of research in population health and health disparities. However, the development and evaluation of measures to capture community characteristics is still at a relatively early stage. The purpose of this work was to assess the reliability of a neighborhood audit instrument administered in the city of Chicago using Google Street View by comparing these “virtual” data to those obtained from an identical instrument administered “in-person”. We find that a virtual audit instrument can provide reliable indicators of recreational facilities, the local food environment, and general land use. However, caution should be exercised when trying to gather more finely detailed observations. Using the internet to conduct a neighborhood audit has the potential to significantly reduce the costs of collecting data objectively and unobtrusively.

DOI PMID

[46]
Kronkvist K.Virtual observations of urban neighborhood physical disorder using Google street view[C]. The Stockholm Criminology Symposium 2014: Program and Abstracts, The Swedish National Council for Crime Prevenion, 2014.

[47]
Mooney S J, Dimaggio C J, Lovasi G S, et al.Use of google street view to assess environmental contributions to Pedestrian injury[J]. American Journal of Public Health, 2016,106(3):462-469.Abstract OBJECTIVES: To demonstrate an information technology-based approach to assess characteristics of streets and intersections associated with injuries that is less costly and time-consuming than location-based studies of pedestrian injury. METHODS: We used imagery captured by Google Street View from 2007 to 2011 to assess 9 characteristics of 532 intersections within New York City. We controlled for estimated pedestrian count and estimated the relation between intersections' characteristics and frequency of injurious collisions. RESULTS: The count of pedestrian injuries at intersections was associated with the presence of marked crosswalks (80% increase; 95% confidence interval [CI]65=652%, 218%), pedestrian signals (156% increase; 95% CI65=6569%, 259%), nearby billboards (42% increase; 95% CI65=657%, 90%), and bus stops (120% increase; 95% CI65=6551%, 220%). Injury incidence per pedestrian was lower at intersections with higher estimated pedestrian volumes. CONCLUSIONS: Consistent with in-person study observations, the information-technology approach found traffic islands, visual advertising, bus stops, and crosswalk infrastructures to be associated with elevated counts of pedestrian injury in New York City. Virtual site visits for pedestrian injury control studies are a viable and informative methodology. (Am J Public Health. Published online ahead of print January 21, 2016: e1-e8. doi:10.2105/AJPH.2015.302978).

DOI PMID

[48]
Gebru T, Krause J, Wang Y, et al.Using deep learning and google street view to estimate the demographic makeup of neighborhoods across the United States[J]. Proceedings of the National Academy of Sciences of the United States of America, 2017,114(50):13108-13113.We show that socioeconomic attributes such as income, race, education, and voting patterns can be inferred from cars detected in Google Street View images using deep learning. Our model works by discovering associations between cars and people. For example, if the number of sedans in a city is higher than the number of pickup trucks, that city is likely to vote for a Democrat in the next presidential election (88% chance); if not, then the city is likely to vote for a Republican (82% chance). The United States spends more than $250 million each year on the American Community Survey (ACS), a labor-intensive door-to-door study that measures statistics relating to race, gender, education, occupation, unemployment, and other demographic factors. Although a comprehensive source of data, the lag between demographic changes and their appearance in the ACS can exceed several years. As digital imagery becomes ubiquitous and machine vision techniques improve, automated data analysis may become an increasingly practical supplement to the ACS. Here, we present a method that estimates socioeconomic characteristics of regions spanning 200 US cities by using 50 million images of street scenes gathered with Google Street View cars. Using deep learning-based computer vision techniques, we determined the make, model, and year of all motor vehicles encountered in particular neighborhoods. Data from this census of motor vehicles, which enumerated 22 million automobiles in total (8% of all automobiles in the United States), were used to accurately estimate income, race, education, and voting patterns at the zip code and precinct level. (The average US precinct contains 1,000 people.) The resulting associations are surprisingly simple and powerful. For instance, if the number of sedans encountered during a drive through a city is higher than the number of pickup trucks, the city is likely to vote for a Democrat during the next presidential election (88% chance); otherwise, it is likely to vote Republican (82%). Our results suggest that automated systems for monitoring demographics may effectively complement labor-intensive approaches, with the potential to measure demographics with fine spatial resolution, in close to real time.

DOI

[49]
Arietta S M, Efros A A, Ramamoorthi R, et al.City forensics: Using visual elements to predict non-visual city attributes[J]. IEEE transactions on visualization and computer graphics, 2014,20(12):2624-2633.We present a method for automatically identifying and validating predictive relationships between the visual appearance of a city and its non-visual attributes (e.g. crime statistics, housing prices, population density etc.). Given a set of street-level images and (location, city-attribute-value) pairs of measurements, we first identify visual elements in the images that are discriminative of the attribute. We then train a predictor by learning a set of weights over these elements using non-linear Support Vector Regression. To perform these operations efficiently, we implement a scalable distributed processing framework that speeds up the main computational bottleneck (extracting visual elements) by an order of magnitude. This speedup allows us to investigate a variety of city attributes across 6 different American cities. We find that indeed there is a predictive relationship between visual elements and a number of city attributes including violent crime rates, theft rates, housing prices, population density, tree presence, graffiti presence, and the perception of danger. We also test human performance for predicting theft based on street-level images and show that our predictor outperforms this baseline with 33% higher accuracy on average. Finally, we present three prototype applications that use our system to (1) define the visual boundary of city neighborhoods, (2) generate walking directions that avoid or seek out exposure to city attributes, and (3) validate user-specified visual elements for prediction.

DOI PMID

[50]
Berland A, Lange D A.Google Street View shows promise for virtual street tree surveys[J]. Urban Forestry & Urban Greening, 2017,21:11-15.Google Street View64 was used to conduct a virtual survey of street trees. The virtual survey performed well for identifying trees to the genus level. Species identification and diameter estimates were less accurate. This approach can generate some street tree data more efficiently than field work.

DOI

[51]
Cheng L, Chu S, Zong W, et al.Use of tencent street view imagery for visual perception of streets[J]. International Journal of Geo-information, 2017,6(9):265.

DOI

[52]
唐婧娴,龙瀛,翟炜,等.街道空间品质的测度变化评价与影响因素识别——基于大规模多时相街景图片的分析[J]. 新建筑,2016(5):110-115.街道空间品质对城市形象和公共生活具有重要影响。试图构建"街道空间品质评估—品质变化特征识别—影响因素分析"的研究框架,利用北京市2005—2013年已有建设用地上的居住类土地出让信息,获取其周边多年份的街景图片,用于评价街道品质和空间变化判读。研究发现,样本街道的总体品质偏低,改善比例在10%左右,且多为表面化的整治。微观环境的优化措施尚未体现精细化设计理念。

DOI

[ Tang J X, Long Y, Zhuo W, et al.Measuring quality of street space, its temporal variation and impact factors: An analysis based on massive street view pictures[J]. New Architecture, 2016,(5):110-115. ]

[53]
Doersch C, Singh S, Gupta A, et al.What makes paris look like paris?[J]. ACM Transactions on Graphics, 2012,31(4):1-9.Given a large repository of geotagged imagery, we seek to automatically find visual elements, e. g. windows, balconies, and street signs, that are most distinctive for a certain geo-spatial area, for example the city of Paris. This is a tremendously difficult task as the visual features distinguishing architectural elements of different places can be very subtle. In addition, we face a hard search problem: given all possible patches in all images, which of them are both frequently occurring and geographically informative? To address these issues, we propose to use a discriminative clustering approach able to take into account the weak geographic supervision. We show that geographically representative image elements can be discovered automatically from Google Street View imagery in a discriminative manner. We demonstrate that these elements are visually interpretable and perceptually geo-informative. The discovered visual elements can also support a variety of computational geography tasks, such as mapping architectural correspondences and influences within and across cities, finding representative elements at different geo-spatial scales, and geographically-informed image retrieval.

DOI

[54]
Cheadle A, Samuels S E, Rauzon S, et al.Approaches to measuring the extent and impact of environmental change in three California community-level obesity prevention initiatives[J]. American Journal of Public Health, 2010,100(11):2129-2136.Abstract Despite growing support among public health researchers and practitioners for environmental approaches to obesity prevention, there is a lack of empirical evidence from intervention studies showing a favorable impact of either increased healthy food availability on healthy eating or changes in the built environment on physical activity. It is therefore critical that we carefully evaluate initiatives targeting the community environment to expand the evidence base for environmental interventions. We describe the approaches used to measure the extent and impact of environmental change in 3 community-level obesity-prevention initiatives in California. We focus on measuring changes in the community environment and assessing the impact of those changes on residents most directly exposed to the interventions.

DOI PMID

[55]
Wilson J S, Kelly C M, Schootman M, et al.Assessing the built environment using omnidirectional imagery[J]. American Journal of Preventive Medicine, 2012,42(2):193-199.Observational audits commonly are used in public health research to collect data on built environment characteristics that affect health-related behaviors and outcomes, including physical activity and weight status. However, implementing in-person field audits can be expensive if observations are needed over large or geographically dispersed areas or at multiple points in time. A reliable and more efficient method for observational audits could facilitate extendibility (i.e., expanded geographic and temporal scope) and lead to more standardized assessment that strengthens the ability to compare results across different regions and studies. The purpose of the current study was to evaluate the degree of agreement between field audits and audits derived from interpretation of three types of omnidirectional imagery. Street segments from St. Louis MO and Indianapolis IN were stratified geographically to ensure representation of neighborhoods with different socioeconomic characteristics in both cities. Audits were conducted in 2008 and 2009 using four methods: field audits, and interpretation of archived imagery, new imagery, and Google Street View鈩 imagery. Agreement between field audits and image-based audits was assessed using observed agreement and the prevalence-adjusted bias-adjusted kappa statistic (PABAK). Data analysis was conducted in 2010. When measuring the agreement between field audits and audits from the different sources of imagery, the mean PABAK statistic for all items on the instrument was 0.78 (archived); 0.80 (new); and 0.81 (Street View imagery), indicating substantial to nearly perfect agreement among methods. It was determined that image-based audits represent a reliable method that can be used in place of field audits to measure several key characteristics of the built environment important to public health research.

DOI PMID

[56]
Naik N, Kominers S D, Raskar R, et al.Computer vision uncovers predictors of physical urban change[J]. Proceedings of the National Academy of Sciences, 2017,114(29):7571-7576.

DOI

[57]
Anderson J M, Macdonald J M, Bluthenthal R, et al.Reducing crime by shaping the built environment with zoning: An empirical study of Los Angeles[J]. University of Pennsylvania Law Review, 2013,161(3):699-756.The authors find that neighborhoods in which there was a zoning change experienced a significant decline in crime.

DOI

[58]
Greenberg S W, Rohe W M, Williams J R.Safety in urban neighborhoods: A comparison of physical characteristics and informal territorial control in high and low crime neighborhoods[J]. Population and Environment, 1982,5(3):141-165.This study addressed the issue of how some urban neighborhoods maintain a relatively low level of crime despite their physical proximity and social similarity to high crime areas. The study explored differences in physical characteristics and various dimensions of the concept of territoriality in three pairs of neighborhoods in Atlanta, Georgia. Neighborhoods within pairs were adjacent and were matched on racial composition and economic status but had distinctly different crime levels. The results indicated that differences in physical characteristics distinguished between matched high and low crime neighborhoods to a far greater extent than did differences in the measures of territoriality. The flow of outsiders into and out of low crime neighborhoods was inhibited because land use was more homogeneously residential, there were fewer major arteries, and boundary streets were less traveled than was the case in high crime neighborhoods. There were relatively few differences in informal territorial control between high and low crime neighborhoods. Where differences existed, informal territorial control was more characteristic of high crime than of low crime neighborhoods.

DOI

[59]
Dimaggio C, Li G.Roadway characteristics and pediatric pedestrian injury[J]. Epidemiologic reviews, 2011,34(1):46-56// Retting R A, Ferguson S A, Mccartt A T. A review of evidence-based traffic engineering measures designed to reduce pedestrian: Motor vehicle crashes[J]. American Journal of Public Health, 2003,93(9):46-56.Abstract Changing the built environment is a sound, but often underutilized approach to injury control. The authors reviewed the literature and conducted a meta-analysis to synthesize the evidence on the association of roadway characteristics with risk of pediatric pedestrian injury. To synthesize the data, they converted results to odds ratios based on direct results or abstracted outcomes and used Bayesian meta-analytic approaches by modeling outcomes as the logit of a normally distributed set of outcomes with vague prior distributions for the central measure of effect and its variance. On the basis of 10 studies of roadway features restricted exclusively to pediatric populations, the synthesized effect estimate for the association of roadway characteristics with pedestrian injury risk was 2.5 (95% credible interval: 1.8, 3.2). The probability of a new study showing an association between the built roadway and pediatric pedestrian injury was nearly 100%. The authors concluded that the built environment is directly related to the risk of pedestrian injury. This review and meta-analysis suggests that even modest interventions to the built roadway environment may result in meaningful reductions in the risk of pediatric pedestrian injury.

DOI PMID

[60]
Forsyth A.What is a walkable place? The walkability debate in urban design[J]. Urban Design International, 2015,20(4):274-292.What exactly is meant by the term ‘walkability’? In professional, research and public debates the term is used to refer to several quite different kinds of phenomena. Some discussions focus on environmental features or means of making walkable environments, including areas being traversable, compact, physically enticing and safe. Others deal with outcomes potentially fostered by such environments, such as making places lively, enhancing sustainable transportation options and inducing exercise. Finally some use the term walkability as a proxy for better design whether composed of multiple, measurable dimensions or providing a holistic solution to urban problems. This review both problematizes the idea of walkability and proposes a conceptual framework distinguishing these definitions. This matters for urban design, because what is considered a walkable place varies substantially between definitions leading to substantially different designs. By mapping the range of definitions, this review highlights potential conflicts been forms of walkability.

DOI

[61]
Frank L D, Schmid T L, Sallis J F, et al.Linking objectively measured physical activity with objectively measured urban form[J]. American Journal of Preventive Medicine, 2005,28(2):117-125.

DOI

[62]
Ewing R, Handy S.Measuring the unmeasurable: Urban design qualities related to walkability[J]. Journal of Urban design, 2009,14(1):65-84.This study attempts to comprehensively and objectively measure subjective qualities of the urban street environment. Using ratings from an expert panel, it was possible to measure five urban design qualities in terms of physical characteristics of streets and their edges: imageability, enclosure, human scale, transparency and complexity. The operational definitions do not always comport with the qualitative definitions, and provide new insights into the nature of these urban design qualities. The immediate purpose of this study is to arm researchers with operational definitions they can use to measure the street environment and test for significant associations with walking behaviour. A validation study is currently underway in New York City. Depending on the outcome of this and other follow-up research, the ultimate purpose would be to inform urban design practice.

DOI

[63]
White J T.Measuring urban design: Metrics for livable places[J]. Journal of Urban Design, 2013,20(2):1-2.No abstract is available for this item.

DOI

[64]
邓小军,王洪刚.绿化率,绿地率,绿视率[J].新建筑,2002,(6):75-76.

[ Deng X J, Wang H G.Green ratio, green space ratio, green looking ratio[J]. New Architecture, 2002,(6):75-76. ]

[65]
赵庆,唐洪辉,魏丹,等.基于绿视率的城市绿道空间绿量可视性特征[J].浙江农林大学学报,2016,33(2):288-294.lt;p>为了解城市绿道植物空间绿量可视性特征，通过构建绿视率计量方法，以广东省立绿道1号线为例进行调查分析，探索了基于绿视率的城市绿道空间绿量可视性特征。结果表明：①省立绿道1号线沿线4个区域总的绿视率均值为31.06%，其中天河区（38.19%）＞海珠区（35.97%）＞越秀区（26.76%）＞荔湾区（26.67%）。②天河区绿道与荔湾区、越秀区绿道绿视率存在显著差异（<em>P</em>＜0.05），表明天河区绿道空间绿量可视性优于荔湾区和越秀区；虽然海珠区绿道绿视率较荔湾区和越秀区高，但通过方差分析多重比较来看，3个区的绿道绿视率没无显著差异（<em>P</em>＞0.05），表明荔湾区、越秀区、海珠区绿道的空间绿量可视性较相似。③越秀区和天河区绿道绿视率较为集中地分布在24%～40%区间，说明越秀区和天河区各自区域绿道内部的绿视率数值较为稳定，其空间绿量可视性效果差异不大，给人感觉一种持续、稳定的绿色空间；而荔湾区和海珠区绿道绿视率的分散在10%~50%，说明荔湾区和海珠区各自区域绿道内部的绿视率数值不稳定，其空间绿量可视性效果差异较大，给人一种间断、具有冲击力的绿色空间。基于绿视率的空间绿量可视性可用于评价城市森林建设的视觉效果。图4表4参21</p>

DOI

[ Zhao Q, Tang H H, Wei D, et al.Spatial visibility of green areas of urban greenway using the green appearance percentage[J]. Journal of Zhejiang A&F University, 2016,33(2):288-294. ]

[66]
Mullaney J, Lucke T, Trueman S J.A review of benefits and challenges in growing street trees in paved urban environments[J]. Landscape and Urban Planning, 2015,134:157-166.Street trees are an integral element of urban life. They provide a vast range of benefits in residential and commercial precincts, and they support healthy communities by providing environmental, economic and social benefits. However, increasing areas of impermeable surface can increase the stresses placed upon urban ecosystems and urban forests. These stresses often lead tree roots to proliferate in sites that provide more-favourable conditions for growth, but where they cause infrastructure damage and pavement uplift. This damage is costly and a variety of preventative measures has been tested to sustain tree health and reduce pavement damage. This review explores a wide range of literature spanning 30 years that demonstrates the benefits provided by street trees, the perceptions of street trees conveyed by urban residents, the costs of pavement damage by tree roots, and some tried and tested measures for preventing pavement damage and improving tree growth.

DOI

[67]
Seiferling I, Naik N, Ratti C, et al.Green streets-Quantifying and mapping urban trees with street-level imagery and computer vision[J]. Landscape and Urban Planning, 2017,165:93-101.Traditional tools to map the distribution of urban green space have been hindered by either high cost and labour inputs or poor spatial resolution given the complex spatial structure of urban landscapes. What more, those tools do not observe the urban landscape from a perspective in which citizens experience a city. We test a novel application of computer vision to quantify urban tree cover at the street-level. We do so by utilizing the open-source image data of city streetscapes that is now abundant (Google Street View). We show that a multi-step computer vision algorithm segments and quantifies the percent of tree cover in streetscape images to a high degree of precision. By then modelling the relationship between neighbouring images along city street segments, we are able to extend this image representation and estimate the amount of perceived tree cover in city streetscapes to a relatively high level of accuracy for an entire city. Though not a replacement for high resolution remote sensing (e.g., aerial LiDAR) or intensive field surveys, the method provides a new multi-feature metric of urban tree cover that quantifies tree presence and distribution from the same viewpoint in which citizens experience and see the urban landscape.

DOI

[68]
Ye Q X, Gao W, Wang W Q, et al.A color image segmentation algorithm by using color and spatial information[J]. Journal of Software, 2004,15(4):522-530.An algorithm for color image segmentation, based on color and spatial information is proposed in this paper. First, color quantization is performed on an image based on the proposed color coarseness metric, and then an incremental region growing method is exploited to find the spatial connectivity of pixels with similar colors to form the initial segmented regions. Second, the initial regions are hierarchically merged based on the region distance defined by the color and spatial information. A criteria is proposed to decide the termination of the merging process. Finally, the erosion and dilation operators are used to smooth the edges of the segmented regions. The experimental results demonstrate that the color image segmentation results of the proposed approach hold favorable consistency in terms of human perception.

DOI

[69]
Zhang L, Gu Z, Li H.SDSP: A novel saliency detection method by combining simple priors[C]. Processing (ICIP), 2013 20th IEEE International Conference, IEEE,2013: 171-175.

Options
Outlines

/

〈 〉