【作者】 黄磊；

【导师】 任福；

【作者基本信息】 武汉大学 ， 地图制图学与地理信息工程， 2019， 硕士

【摘要】 随着城市化进程的加快,城市的宜居性问题越来越受到人们的重视。街道作为城市居民生活的聚集地,也是居民与城市环境发生交互的窗口,街道的局部场景深刻影响着居民在城市活动中的体验和身心健康,如何定量评估人们对于街道局部场景的感知,一直是相关研究领域重点关注的问题。传统的方法受限于样本的数量和数据处理手段,难以开展大范围内的场景感知研究。随着街景数据的可获得性大大提升以及计算机视觉技术的日臻成熟,为城市场景感知研究提供了新思路。本文以基于百度街景数据,通过SegNet语义分割框架将场景具体划分为天空、建筑、树等12个类别,结合景观生态学多样性理论,计算基于街景的景观丰富度指数、Shannon-Weaver指数、Simpson指数等作为街道场景多样评价指标;从街景中提取绿化率、视觉开放性、视觉封闭性指数作为街道场景视觉质量评价指标;另外将机动化程度、汽车出现的概率、行人和自行车出现的概率作为辅助评价指标;基于POI和道路网数据计算功能混合度、重要功能POI密度、道路网密度等指标。通过以上四类指标构建场景感知计算模型,用以量化人们对局部场景的感知。利用两周的腾讯宜出行定位数据表征深圳市人群活动的时空特征。基于场景感知计算模型和城市人群活动的时空特征,通过时空地理加权回归模型分析场景感知与城市人群活动之间的关系,从而更好地理解人们对于局部场景的感知以及场景感知要素对城市人群活动的影响。研究发现,场景感知计算模型能够较好地定量评估人们对于局部场景的感知,场景感知要素对城市人群活动的影响是随着时间和空间的变化而变化的。与直接使用分割要素所占比例的量化方式相比,本研究提供了一种更具解释性的场景感知量化方法,有助于研究人员理解城市基础结构,揭示城市功能对人的行为的影响。更多还原

【Abstract】 With the rapid development of urbanization,tough attention has been paid to urban livability.Street plays an important role in physical activity and urban life,which have a particularly profound influence on the experience of physical activity and health.Understanding and quantifying the human perceptions of locale environment have long been of hot to a wide variety of fields.Previous studies are subject to limited samples and inefficient means of data processing,which is hard to measure human perceptions of locale environment for a large-scale urban area.Currently,some new ideas for human scene perceptions come into being with the increasing availability of street view and cutting-edge computer vision.Based on Baidu Street View image covering the Shenzhen city,the classification of 12 road scene categories is achieved under the help of SegNet that is a semantics segmentation framework,such as sky,building,tree and so on.Subsequently,the diversity theory of Landscape Ecology is employed to measure the variety of street scene,namely,Richness Index,Shannon-Weaver Index and Simpson Index for street view.Besides,Green View Index,the Openness and Enclose of vision are proposed to illustrate visual quality of street space.In addition,the motorization of street,the radio of car,the radio of pedestrian and bike are taken as the representative elements.Finally,diversity of street’s function,density of typical POI and road density are extracted from Points of Interest(POI)and road networks provided by Open Street Map(OSM).Consequently,a model for scene perceptions is provided on the basis of four indicators mentioned above,aiming to quantify human perceptions for street scene.In the meanwhile,the spatio-temporal distribution of human activities in Shenzhen is mapping by integrating Tencent user density over two weeks.Furthermore,in consideration of spatial and temporal non-stationarity of data,Geographically and Temporally Weighted Regression(GTWR)is used to explore the relationships between human perceptions for street scene and the spatio-temporal distribution characteristics of urban population.The results show that the model of scene perceptions could partly quantify human perceptions for street scene,and the spatio-temporal distribution of urban population is influenced by various factors that are heterogeneous over space and time.Compared with traditional methods of employing the proportion of segmentation elements,this study arms researchers with more objective measures of human perceptions for street scene,assisting researchers with understanding the underlying urban structure and revealing the impacts of urban function.更多还原