街景影像在城市交通研究中的应用:回顾、分析和展望

The use of street-view images in urban traffic studies:review,analysis,and outlook

街景影像覆盖面广,能提供城市级别的交通场景信息,对开展交通研究分析提供了大规模数据源的支持和新的研究方法。为了探究街景影像在交通研究中的应用情况,从Web of Science核心集合数据库筛选了2011—2022年街景图像在交通研究中应用相关的143篇论文,借助CiteSpace文献计量分析软件从年发文量、作者合作图谱、国家与机构合作图谱、关键词共现、关键词聚类和主题突发检测等方面进行归纳分析。在此基础上总结街景影像在交通基础设施、交通安全感知、出行辅助和出行环境感知四方面的应用研究进展,并对未来的研究方向提出展望。文献综述结果表明:(1)街景影像数据已被广泛应用于交通领域不同维度的研究,大多数研究通过卷积神经网络模型提取街景影像信息以反映交通场景特征;(2)由于街景数据采集时间跨度大,致使当前基于街景影像数据的交通方面应用主要集中在空间维度研究,缺乏动态时间维度的分析;(3)街景影像与交通领域知识数据进行融合分析建模是街景影像数据在交通领域应用的发展趋势。

Street-view images encompass a wide range of areas and offer traffic insights at the city level,supporting large-scale data sources and new methods for traffic research and analysis.To investigate the application of streetscape images in traffic research,143 papers related to this topic from the Web of Science core collection database spanning from 2011 to 2022 were screened.These papers were analyzed utilizing CiteSpace bibliometric analysis software,focusing on parameters such as annual publication volume,author cooperation map,country and institution cooperation map,keyword co-occurrence,keyword clustering,and theme burst detection.Based on this,we summarize advancements in research concerning the application of street-view images across four aspects:transportation infrastructure,traffic safety perception,travel assistance,and travel environment perception.Additionally,we offer insights into prospective avenues for future research.The results of the literature review show that:(1)street-view image data have been widely used across different dimensions of the transportation sector,with a predominant utilization of convolutional neural network models to extract traffic scene characteristics;(2)despite the extensive temporal scope of street-view data collection,current traffic applications mainly focus on the spatial dimension of analysis,neglecting the dynamic time dimension;and(3)a notable emerging trend involves the fusion analysis and modeling of street-view images with traffic knowledge data,aiming to enhance their application within the transportation domain.