基于地理标记全景图的城市街道可视空间分析

Visible Street Space Analysis Based on Geo-tagged Panoramic Images

地理标记全景图指带有地理位置信息的全景图片。以武汉市建成区地面街道为研究对象,以OSM路网数据和百度街景全景图为数据来源,将百度街景的原始等距全景转换为等积圆柱投影图像,基于DRN-D-105神经网络模型,针对城市街道可视空间的4项可量化指标,天空可视率、植被可视率、建筑可视率和道路可视率的空间分布进行评估。并针对武汉市建成区不同街道类型、不同街道区域进行分析对比。同时将自动计算结果同人工识别结果计算IoU进行比较,评估了神经网络模型计算各项指标的准确度。

A geo-tagged panorama is a panoramic photo labeled with geographic location information.The ground streets in the built-up area of Wuhan are selected as the research area.The OSM road data and Baidu Street View panoramic photos are downloaded as the data source.Then the original equirectangular projected panoramic photos of Baidu Street View are converted into equal-area cylindrical projected images.The DRN-D-105neural network model is used for semantic segmentation to calculate four quantifiable indicators of urban street visible space:visible sky rate,visible green rate,visible building rate,and visible pavement rate.Then the values by different street types and different street areas in the built-up area of Wuhan City were carried out for analysis and comparison.Furthermore,the automatic recognition results are compared with the manual recognition results by calculating the IoU to evaluate the semantic segmentation accuracy.