众包时空数据驱动的城市地理信息推测综述

Knowing the accurate geographic information is the basis to achieve the intelligent decisions in cities. Tra⁃ ditional geographic information collecting mainly relies on manual mapping, patrolling or sensing by static geographical sensors, which are expensive given specialized equipment and labors. Recently, with the development of the mobile Inter⁃ net, the ubiquitous moving objects has generated massive spatio-temporal data in the urban spaces, who act as sensors of the city consciously or unconsciously, and make it possible to infer the geographic information based on those data in a crowdsourced manner. The geographic information inference based on the crowd-sourced spatio-temporal data enjoys the advan⁃ tages of low cost, high spatial coverage, and timely updates. However, it also has the data quality issues, which introduce great challenges to the urban geographic information inference. In this paper, we survey the location and attribute inference of geospatial entities, including the road network, point of interest and area of interest, based on crowd-sourced spatio-tem⁃ poral data, e.g., trajectories, location-based social network, and street views. We give the definitions of crowd-sourced spa⁃ tio-temporal data and geospatial entities, compare the pros and cons of utilizing the crowd-sourced spatio-temporal data to infer the geographic information against traditional methods, and elaborate the research problems and challenges. After that, we review four research problems, i.e., map matching, name extraction, location discovery and statistical attribute infer⁃ ence. Finally, we present the future research directions and conclude the paper.