Changes of Street Quality Based on Street View Images in Different Periods:A Case Study of Jingshan East Street,Dongcheng District,Beijing City

With the continuous calls for energy conservation and emission reduction in recent years,more and more people choose walking as their travel mode.The improvement of the quality of street space will directly affect people's willingness to walk.By sorting out relevant research on street quality measurement,extracting quality keywords with high frequency of reference as impact factors,and using street view image data from different eras,semantic segmentation technology,factor analysis,and questionnaire survey methods,this paper evaluates the street quality of Jingshan East Street,Dongcheng District,Beijing,further explores the impact of different factors on street quality,and analyzes possible ways to improve it.

Rapid visual screening of soft-story buildings from street view images using deep learning classification

Rapid and accurate identification of potential structural deficiencies is a crucial task in evaluating seismic vulnerability of large building inventories in a region. In the case of multi-story structures, abrupt vertical variations of story stiffness are known to significantly increase the likelihood of collapse during moderate or severe earthquakes. Identifying and retrofitting buildings with such irregularities—generally termed as soft-story buildings—is, therefore, vital in earthquake preparedness and loss mitigation efforts. Soft-story building identification through conventional means is a labor-intensive and time-consuming process. In this study, an automated procedure was devised based on deep learning techniques for identifying soft-story buildings from street-view images at a regional scale. A database containing a large number of building images and a semi-automated image labeling approach that effectively annotates new database entries was developed for developing the deep learning model. Extensive computational experiments were carried out to examine the effectiveness of the proposed procedure, and to gain insights into automated soft-story building identification.

基于Photo Sphere Viewer的轻量化可量测街景系统研究

随着新型基础测绘体系发展,测绘技术和装备不断升级,大量街景数据随着车载激光扫描工作一并被采集。针对该数据路线随机性大、密集程度高等特点,以及传统街景地图系统插件受限、扩展性不强、无法量测等局限,本文设计了基于Photo Sphere Viewer的轻量化可量测街景系统,在开源插件的基础上实现了对其组件的扩充、实现了行进方向、地图交互等系统功能的设计优化,进一步实现了基于深度图的街景地图量测功能。生产作业队伍对系统的使用结果表明,该系统可以辅助生产作业队伍进行属性采集、纹理更新、数据检查等多项工作,操作灵活便捷,且具有较强的可扩展性,便于基于生产需求进行改进调整,提高了生产效率,为新型基础测绘成果进一步深化应用提供了良好的借鉴作用。

Panorama completion for street views

This paper considers panorama images used for street views. Their viewing angle of 360° causes pixels at the top and bottom to appear stretched and warped. Although current image completion algorithms work well, they cannot be directly used in the presence of such distortions found in panoramas of street views. We thus propose a novel approach to complete such 360° panoramas using optimizationbased projection to deal with distortions. Experimental results show that our approach is efficient and provides an improvement over standard image completion algorithms.

Local color and morphological image feature based vegetation identification and its application to human environment street view vegetation mapping, or how green is our county?

Measuring the amount of vegetation in a given area on a large scale has long been accomplished using satellite and aerial imaging systems.These methods have been very reliable in measuring vegetation coverage accurately at the top of the canopy,but their capabilities are limited when it comes to identifying green vegetation located beneath the canopy cover.Measuring the amount of urban and suburban vegetation along a street network that is partially beneath the canopy has recently been introduced with the use of Google Street View(GSV)images,made accessible by the Google Street View Image API.Analyzing green vegetation through the use of GSV images can provide a comprehensive representation of the amount of green vegetation found within geographical regions of higher population density,and it facilitates an analysis performed at the street-level.In this paper we propose a fine-tuned color based image filtering and segmentation technique and we use it to define and map an urban green environment index.We deployed this image processing method and,using GSV images as a high-resolution GIS data source,we computed and mapped the green index of Milwaukee County,a 3,082 km^(2) urban/suburban county in Wisconsin.This approach generates a high-resolution street-level vegetation estimate that may prove valuable in urban planning and management,as well as for researchers investigating the correlation between environmental factors and human health outcomes.

Exploring associations between streetscape factors and crime behaviors using Google Street View images

Understanding the influencing mechanism of the urban streetscape on crime is fairly important to crime prevention and urban management.Recently,the development of deep learning technology and big data of street view images,makes it possible to quantitatively explore the relationship between streetscape and crime.This study computed eight streetscape indexes of the street built environment using Google Street View images firstly.Then,the association between the eight indexes and recorded crime events was revealed with a poisson regression model and a geographically weighted poisson regression model.An experiment was conducted in downtown and uptown Manhattan,New York.Global regression results show that the influences of Motorization Index on crimes are significant and positive,while the effects of the Light View Index and Green View Index on crimes depend heavily on the socioeconomic factors.From a local perspective,the Pedestrian Space Index,Green View Index,Light View Index and Motorization Index have a significant spatial influence on crimes,while the same visual streetscape factors have different effects on different streets due to the combination differences of socioeconomic,cultural and streetscape elements.The key streetscape elements of a given street that affect a specific criminal activity can be identified according to the strength of the association.The results provide both theoretical and practical implications for crime theories and crime prevention efforts.

Charting Disaster Recovery via Google Street View: A Social Science Perspective on Challenges Raised by the Fukushima Nuclear Disaster

There is increasing interest in using Google Street View(GSV) for research purposes, particularly with regard to ‘‘virtually auditing'' the built environment to assess environmental quality. Research in this field to date generally suggests GSV is a reliable means of understanding the ‘‘real world'' environment. But limitations around the dates and resolution of images have been identified. An emerging strand within this literature is also concerned with the potential of GSV to understand recovery post-disaster. Using the GSV data set for the evacuated area around the Fukushima Dai'ichi nuclear power plant as a case study, this article evaluates GSV as a means of assessing disaster recovery in a dynamic situation with remaining uncertainty and a significant value and emotive dimension. The article suggests that GSV does have value in giving a high-level overview of the postdisaster situation and has potential to track recovery and resettlement over time. Drawing on social science literature relating to Fukushima, and disasters more widely, the article also argues it is imperative for researchers using GSV to reflect carefully on the wider socio-cultural contexts that are often not represented in the photo montage.

A news picture geo-localization pipeline based on deep learning and street view images

Numerous news or event pictures are taken and shared on the internet every day that have abundant information worth being mined,but only a small fraction of them are geotagged.The visual content of the news image hints at clues of the geographical location because they are usually taken at the site of the incident,which provides a prerequisite for geo-localization.This paper proposes an automated pipeline based on deep learning for the geo-localization of news pictures in a large-scale urban environment using geotagged street view images as a reference dataset.The approach obtains location information by constructing an attention-based feature extraction network.Then,the image features are aggregated,and the candidate street view image results are retrieved by the selective matching kernel function.Finally,the coordinates of the news images are estimated by the kernel density prediction method.The pipeline is tested in the news pictures in Hong Kong.In the comparison experiments,the proposed pipeline shows stable performance and generalizability in the large-scale urban environment.In addition,the performance analysis of components in the pipeline shows the ability to recognize localization features of partial areas in pictures and the effectiveness of the proposed solution in news picture geo-localization.

全景影像在城市研究中的应用进展综述

全景成像技术的进步,街景图像工具的普及,以及人工智能领域的计算机视觉、机器学习和深度学习技术的快速发展,推动了在城市研究中利用全景影像进行大规模、自动化的判别与解析。上述领域的快速发展促使近20年来全景影像、人工智能和城市研究领域之间涌现了大量交叉成果。借助文献计量工具中常用的CiteSpace和VOSviewer作为分析平台,梳理了全景影像在城市研究中的应用进展。首先利用文献共被引聚类网络与术语时区图,划分了全景影像在城市研究中的三个发展阶段。然后借助合著网络和关键词聚类分析,梳理了各阶段全景影像在城市研究中的合著关系、全景影像的获取方式、图像信息的提取技术,归纳了全景影像在城市研究中的四个主要应用领域:城市建成环境、城市景观环境、城市物理环境和智慧城市。最后在历史分期视域下,剖析了促成全景影像应用领域发展的主要驱动因素,并总结了应用全景影像的城市研究目前存在的挑战和未来的发展趋势。

基于机器学习和街景图像的历史城区店招色彩量化研究——以苏州为例

店招色彩是城市风貌的重要组成元素,特别是风貌重点管控的历史城区。文章运用机器学习技术和街景图像数据,以苏州历史城区为例,基于“传统建筑一管控建筑一现代建筑”3个尺度构建一套历史城区店招色彩的评价方法。该方法以街道为单位识别不同类型建筑的店招主导色彩并对街道的店招色彩协调度进行评价。通过筛选协调度较高和较低的街道,分析其店招色彩特点,提出历史城区店招色彩量化管控的依据。该方法可为历史城区店招的精细化更新管控提供技术支持。

深度学习支持下的城市破损路沿石检测方法

针对破损路沿石在街景影像中受到目标多尺度、相似地物干扰以及遮挡等问题,提出了一种面向城市街道两侧破损路沿石检测的CDD-YOLO(convolutional swin transformer deformable decouple-YOLO,CDD-YOLO)模型。依据破损路沿石呈现形状尺度多样性特点,嵌入C3_STR(convolutional swin transformer,C3_STR)模块进行特征融合,增强模型对多尺度特征的感知性能;对于相似地物干扰导致的误检现象,加入可变形卷积模块,利用目标区域自适应特性,提升模型对相似地物的判别能力;为避免因遮挡引起的定位不准确问题,引入解耦检测头结构,增强模型对模糊边界特征的提取能力。在自制的街景破损路沿石数据集上进行验证,分析表明,该方法的precision、recall、F1、IoU 4项评价指标分别达到了82.45%、81.22%、81.01%和80.23%,显著优于其他主流目标检测方法,验证了该方法的有效性和可行性。

基于全景图的城市街谷平均辐射温度计算模型研究

在评估城市街谷热环境对居民生活质量的影响时,明确街谷的微气候条件,尤其是平均辐射温度(Mean Radiant Temperature,MRT)对于改善城市热环境具有重要意义。现阶段,利用基于鱼眼图的天空视域因子(Sky View Factor,SVF)计算方法较为繁琐且难以实现大范围街谷MRT的时空分布评估。因此,本研究旨在通过全景图像技术,提出一种快速大量计算城市街谷MRT时空分布的新方法,并进一步考虑城市街谷中树木的影响,以改进传统模型。本研究首先基于全景图像批量获取城市街谷中的SVF,并结合城市街谷中的几何特征和植被视域因子,通过改进的平均辐射温度计算模型对单点的MRT进行计算,同时采用定点实测数据对该模型进行精确度验证,并将其应用于西安市街谷MRT的实际计算中。研究结果显示,经过模型验证,本研究方法具有较高的精度,相对误差大多数情况下在20%以内,RMSE在2.85~4.66℃。同时,模型与实测数据的一致性较好,能够清晰地反映MRT的变化趋势,R2大于0.74,IA大于0.80。与仅考虑单一不透水面的计算模型相比,考虑树木植被后的模型精度有明显提升,RMSE从5.15℃降至3.87℃,模型的R2由0.72提升至0.74,表明改进模型与观测结果具有更好的一致性。提出的基于全景图的MRT计算方法不仅提高了评估城市街谷热环境的效率和精度,而且通过考虑树木植被的影响,为城市规划和绿化管理提供了更加科学的指导。此外,本研究的方法和结论能够为城市热岛效应的缓解和城市生活环境的改善提供理论依据和技术支持。通过实例分析,研究成功地应用于西安市,展示了2021年7月14日上午9:00的街谷MRT分布图,为后续的城市热环境评估和改善工作奠定了基础。

基于跨层次聚合网络的实时城市街景语义分割

随着自动驾驶技术的迅速发展,精确高效的场景理解显得尤为重要。城市街景语义分割旨在准确识别并分割出行人、障碍物、道路和标志物等要素,为自动驾驶技术提供必要的道路信息。然而,当前的语义分割算法在城市街景分割中仍然面临一些挑战,主要表现为不同类别的像素区分不够清晰、对于复杂场景结构的理解不够精准以及对小尺度对象或大尺度结构的分割不准确等问题。为此,本文提出一种基于跨层次聚合网络的实时城市街景语义分割算法。首先,在编码器末端设计了结合跨层次聚合的金字塔池化模块,用于高效提取多尺度上下文信息;其次,在编码器和解码器之间设计了跨层次聚合模块,通过引入通道注意力机制增强信息的表征能力,逐级聚合编码器阶段的特征以充分实现特征复用;最后,在解码器阶段设计了多尺度融合模块,在通道维度聚合全局信息与局部信息,促进深层特征与浅层特征的融合。将所提算法在两个通用的城市街景数据集上进行了验证。在一张RTX3090显卡上(TensorRT测速环境),本文算法在Cityscapes测试集以294 FPS的实时性达到73.0%mIoU的准确性,在更高分辨率的图像上以164 FPS的实时性达到75.8%mIoU的准确性;在CamVid数据集以239 FPS的实时性达到74.8%mIoU的准确性。实验结果表明,本文算法在准确性与实时性之间取得了有效平衡,对比其他算法的语义分割性能具有显著提升,为实时城市街景语义分割领域带来了新的突破。

利用DeepLabv3+模型提取分析街景图像绿视率——以北京三环内为例

基于语义分割模型的绿视率提取缺乏适用性研究,本文首先基于DeepLabv3+语义分割预训练模型和自主标注样本,采用迁移学习策略,构建街景图像语义分割模型,并对其进行精度验证。然后基于构建的街景图像语义分割模型提取计算北京三环内绿视率(GVI),分析点、线尺度下绿视率空间分布特征。结果表明:①相比DeepLabv3+语义分割预训练模型,迁移学习后模型F1值和mIoU值分别提高了7%和3%;②点状尺度上北京三环内绿视率整体呈“西高东低,北高南低”聚类式分布特征,0~0.15区间内街景采样点GVI约占58.1%;③线状尺度上整体呈“环线低环内高”且中心发散式特征分布,0~0.15区间内研究区道路GVI约占59.8%。该研究对于提升城市街道绿化感知程度和城市空间规划具有重要的参考意义。

一种街景图像中建筑物高度估算方法

建筑物高度信息是城市三维建模的基础数据,但已有的建筑物高度估算研究多采用LiDAR和SAR等遥感影像。随着计算机和互联网的快速发展,街景数据因采集容易和成本低等特点成为了一种新兴的建筑物高度估算数据源。文章提出一种街景图像中建筑物高度估算方法,首先利用segment anything model实现图像中建筑物像素高度提取;然后利用图像元数据和电子地图数据获取建筑物与相机之间的距离、图像焦距,根据街景图像与建筑物实体的几何关系改进针孔相机模型,构建建筑物高度估算方法;最后选取北京、柏林的Mapillary街景图像开展实验验证。结果表明,与改进前相比,改进后针孔相机模型明显提升了高度估算准确度,RMSE降低了11.31 m,R^(2)提高了0.4,具备实用价值。

基于绿视率和NDVI的城市街道景观分析与优化研究

街道景观空间对市民健康和城市风貌具有重要影响。既往研究中常以归一化植被指数(NDVI)和绿视率(GVI)来分别代表二维和三维的绿色指标,但对二者的指标相关性研究甚少。采用基于深度学习的图像语义分割方法分析百度街景计算代表性街道的GVI,利用GF-1卫星数据计算NDVI,比较分析城市街道的GVI和NDVI指标特征及相关性。结果表明,1)中山市中心城区各代表街道GVI指标参差不齐,从8.06%到36.00%,其中石岐街道兴中道GVI最高;2)各街道观测点的NDVI均值随着缓冲区尺度的增加也随之呈现出不同变化,NDVI均值具有强烈的尺度敏感性;3)50 m GVI和DNVI均值的皮尔逊相关系数最高,达到0.832。在此基础上分析街道景观存在的不足并给出优化建议,为城市街景评估、空间优化、景观提升提供参考。

顾及多模态数据特征关联的城市功能区识别方法——以武汉市为例

城市功能区在城市化建设当中扮演着重要的角色,城市功能区建模与分析对于规划城市土地布局和优化土地利用模式有着重要意义。目前在城市功能区建模与分析的研究中,仍存在多模态数据利用率低、关联特征损失大的问题,这导致难以精确捕捉城市空间结构信息。因此,本研究以武汉市中心区域为例,基于街景影像数据和兴趣点数据构建城市功能区识别方法。首先,将研究区域划分为500 m×500 m的格网单元,运用自然语言模型、深度学习、空间插值等方法来提取多维度的城市功能特征;其次,利用核典型相关分析方法提取多模态数据的关联特征,分析不同共享空间维度和相关性的关系;最后,基于建成环境特征、社会经济语义特征、拼接特征及关联特征的多维度特征构建随机森林功能区识别模型,并分别与两种常用数据模态进行对比评价。结果表明,本方法相较于没有顾及关联特征的模型准确度提高了3.4%。研究成果可为政府决策规划、城市发展评估提供帮助,推动城市化建设。

基于城市街景和深度学习的老龄人群安全感知评价研究--以沈阳市中心城区为例

随着我国社会老龄化不断加剧,面向老龄人群的精细化规划设计方法需求迫切。相关研究表明,建成环境的感受是否安全,将直接影响老龄人群开展日常活动的意愿和强度。在新技术的支持下,大规模快速测度城市空间情绪感知已成为可能。因此,建立老龄人群的城市空间“感知—评价—优化”分析框架有助于精细化规划设计的实现。本研究以重度老龄化的特大城市——沈阳市中心城区作为研究对象,采用深度学习对城市街景要素进行语义分割;建立人机对抗的老龄人群安全感知随机森林模型,并对沈阳中心城区的街景进行评价。此外,研究还采用冷热点聚类、增强归回树(BRT)模型等方法,探讨了城市空间中老龄人群安全感知与建成环境之间的关系。结果表明:第一,沈阳市老龄安全感知水平呈现显著的空间异质性,总体趋势为内高外低;第二,研究基于冷热点聚类模型分析得到了安全感知空间聚类特征,并分析其与不同用地性质的概率分布关系;第三,在不同用地类型下城市老龄安全感知冷热点集聚差异显著,居住用地热点呈片状分布、商业用地呈南北线状分布、公共服务设施用地由于分散且数量较少而聚类不显著;第四,研究基于BRT模型分析了人机对抗模型中,街景要素与安全感知水平的贡献水平和边际效应。研究结果对面向老龄人群的精细化规划和设计提供了一定支撑。

基于街景和POI数据的街道空间品质评价

街道空间是居民生活的主要场所之一,其空间品质的优劣程度不仅是城市居民幸福感的体现,也是反映城市形象的重要因素。以渝中区道路为研究对象,利用图像语义分割技术、空间句法及空间连接等方法,分别从街景图像、路网和POI数据中获得绿视率、道路接近度、太阳辐射强度、生活服务便捷度等13个评价指标,基于层次分析法构建包含舒适度、安全性、便捷性和活力度4个感知维度的街道空间品质评价体系。最终分别对整体4个感知维度及道路等级的街道空间品质分布特征进行分析,结果表明渝中区各街道空间品质多样性与差异性并存,且未出现明显的两极分化现象。

基于SE-Mask-RCNN建筑遗产识别与空间可视化分析

传统建筑是中国宝贵的建筑遗产,承载着优秀的民族建筑文化,是反映城市特色风貌的重要指标。现阶段深度学习识别建筑物的技术相对成熟,但使用街景图片识别建筑遗产并进行地图可视化展示的研究较少,因此,本研究基于Mask-RCNN(mask region-based convolutional neural network)模型,融合SE(squeeze and excitation)注意力机制,提出一种基于SE-Mask-RCNN识别街景图片中建筑遗产的方法。首先,通过路网数据获取百度街景图片,制作数据集。其次,在模型的残差网络(residual network,ResNet)中引入SE注意力机制;并与已有相关方法 U-net(u-shaped network)、全卷积网络(fully convolutional network,FCN)、Mask-RCNN三种模型进行实验对比评价。最后,使用本方法识别研究区域内的街景图片,形成可视化地图,分析建筑遗产在空间上的分布情况。结果表明,本方法可以有效识别城市中的建筑遗产,识别结果较Mask-RCNN、U-Net、FCN模型分别提高了2%、3.1%、4.7%,证明了本方法对城市中建筑遗产的识别具有可靠性和有效性。研究成果可为建筑遗产保护及现状调查提供依据。