Abstract
The number of applications associated with OpenStreetMap (OSM), one of the most famous crowd-sourced projects for volunteered geographic information (VGI), have increased because OSM data is both `free' and `up-to-date'. However, limited by the ability of the providers, the quality of the collected data remains a valid concern. This work focuses on how to assess the quality of OSM via deep learning and high-resolution remote imagery. First, considering that high-resolution remote sensing imagery is clear enough for recognizing buildings, we proposed using multi-task deep-convolutional networks to extract buildings in pixel level. The extracted buildings were converted into polygons with geographical coordinates, which were treated as reference data. Then, OSM footprint data were matched with the reference data. Quality was assessed in terms of both positional accuracy and data completeness. Finally, the building footprint data of OSM for the city of Las Vegas, Nevada, USA, were evaluated. The experiments show that the proposed method can assess OSM effectively and accurately. The results show that building extracted by the proposed method can be treated as a new data source for assessing OSM quality and can also be used for urban planning in regions where OSM lacks building data.
Highlights
Due to the development of Web 2.0 and sensors, an increasing number of crowd-sourced projects are being applied by users, which involves user-generated content (UGC)
To solve the above problems—considering that highresolution remote sensing imagery is easier to obtain than authoritative vector data—and bolstered by the encouraging performance of deep convolutional networks when extracting buildings from images, we propose a new method for assessing the quality of OSM building footprint data using high-resolution remote sensing imagery
All the images and ground truth data were clipped into 8,086 cells with 650 × 650 pixels (Figure 6). This experiment split all the data into two parts, where eighty percent were treated as the training dataset and used to train the deep neutral network, and the remaining twenty percent were treated as test dataset and used to evaluate the generated polygons and confirm the building footprints
Summary
Due to the development of Web 2.0 and sensors, an increasing number of crowd-sourced projects are being applied by users, which involves user-generated content (UGC). The use of user-generated geographic data (e.g., volunteered geospatial information; VGI [1] has increased substantially in geographic information science (GIS). Most countries have a national mapping or cadastral agency to collect geospatial data. Limited by the condition that only a few of these collected geospatial data are accessible by the public, the crowd-sourced project OpenStreetMap (OSM) has become increasingly famous. It has been used to optimize freight transport in route planning systems [2], refining building-level population estimates via multi-class dasymetric mapping and interpolation by surface volume integration [3], generating land-use maps [4] and so on.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
