基于多源大数据与深度学习的道路视觉 特征与骑行流量相关性研究

伴随着市民慢行需求的不断增长，在国家的大力倡 导下，城市慢行系统建设得到迅速发展。而现有关于城市慢行 系统的研究中，针对道路视觉特征与骑行流量的相关性研究较 少，骑行道路的建设预评价缺乏科学的数据支撑。以北京市海 淀区5类可供骑行的市政道路为研究对象，基于全景静态图数 据，结合深度学习方法对道路的绿景指数、街道开敞度、建筑 比及界面围合度4项视觉特征进行量化，并利用开源运动数据 Heatmap获取道路对应的骑行流量，对视觉特征与骑行流量 之间的关系展开相关性研究。研究发现：在各类道路中，绿景 指数与建筑比均与骑行流量呈正相关，而随市政道路级别升 高，街道开敞度则与骑行流量由正相关逐渐转为负相关。研究 结果可以为基于骑行使用的道路建设或改造及相关绿道规划设 计提供依据。

Research on the Correlation between Road Visual Features and Cycling Traffic Based on Multi-source Big Data and Deep Learning

With the continuous growth of citizens'' demand for slow travel, the construction of urban slow travel system has developed rapidly under the vigorous advocacy in Chia. In the existing research on urban slow-moving systems, there are few studies on the correlation between road visual characteristics and cycling traffic, and the pre-evaluation of cycling roads lacks scientific data support. This paper takes five types of municipal roads available for cycling in Haidian District, Beijing as the research object. Based on panoramic static map data, combined with deep learning methods, the four visual features of the road, including green scene index, street openness, building ratio and interface enclosure, were quantified, and the open-source motion data Heatmap was used to obtain the corresponding cycling traffic of the road. A correlation study was carried out on the relationship between visual features and cycling traffic. The study found that in all types of roads, the green landscape index and building ratio were positively correlated with cycling flow; while with the increase of municipal road level, street openness and cycling flow gradually turned from positive correlation to negative correlation. The results of this study can provide a basis for road construction or renovation based on cycling use and related greenway planning and design.