基于多源GPS轨迹与社交媒体数据的城市道路图自动提取

Extracting and updating road information within cities from user generated trajectories has become a hotspot in the field of geographic information science. However, the existing methods are mainly suitable for single-source GPS trajectory data, which ignores the heterogeneity characteristics of user generated trajectories in sampling frequency and positioning precision, and no existing work has provided a solution to automatically extract the attribute information of roads. In view of the above problems, this research aims at proposing an automatic extraction method of city road maps based on multisource GPS trajectory and social media data. On one hand, we develops a membership distribution model representing the coupling patterns between GPS trajectories and road region, based on which the center lines of road lanes can be extracted. Furthermore, we use the support vector machine to select the optimum feature sets for road grade inferring, and finally reconstruct hierarchical road networks. On the other hand, we create a large-scale social media corpus using the established description rules for named entity of roads, further train the condition random field model to recognize the named entities of roads, and at last match name information with road segments. The proposed method can automatically extract city road maps from low-cost and short-update-cycle crowdsourcing data and has a promising prospect of supplementing the existing map information collection modes.

利用“维基”理念借助公众分享轨迹数据提取与更新城市道路信息已经成为地理信息科学领域的热点问题，但现有方法主要针对单一来源GPS轨迹数据，未顾及公众分享数据采样频率多样、定位精度不一等异质性问题，且鲜有研究提供道路属性信息自动提取的解决方案。针对上述问题，本项目旨在提出基于多源GPS轨迹与社交媒体数据的城市道路图自动提取方法，一方面从统计角度发展表达GPS轨迹与道路区域耦合规律的隶属度分布模型，以模型为基础提取车道中心线，并采用支持向量机选取最优特征组合推理道路等级，重构车道级的层次道路网；另一方面通过构建道路命名实体描述规则，建立大规模社交媒体语料库，进而训练条件随机场模型，自动识别社交媒体文本中的道路命名信息，并实现路段与命名信息匹配。本研究通过低成本、更新快的众源数据自动提取城市道路图，有望成为现有信息采集方式的良好补充，服务于基础地理空间框架建设。

利用“维基”理念借助公众分享轨迹数据提取与更新城市道路信息已经成为地理信息科学领域的热点问题，但现有方法未顾及公众分享数据采样频率多样、定位精度不一等异质性问题，且鲜有研究提供道路属性信息自动提取的解决方案。针对上述问题，本项目实现了基于GPS轨迹与社交媒体数据的城市道路图自动提取方法，并研究了基于遥感影像和GIS数据的道路提取方法；另一方面通过构建道路命名实体描述规则，建立社交媒体语料库，自动识别社交媒体文本中的道路命名信息，实现了路段与命名信息匹配；最后，开展了城市数字道路图自动提取与挖掘应用。项目成果能够从低成本、更新快的众源数据自动提取城市道路图。