Geospatial street-view retrieval methodology using geo-tagged photos

Recently, a generation of photo datasets becomes easier on internet and cloud services. However, conventional point based photo retrieval would be often irritable for users in a disaster monitoring and construction management. Therefore, an assistance of photo retrieval is required to improve efficiency in the monitoring and guidance. In this study, we focus on an improvement of photo browsing using geo-tag such as camera position and azimuth data taken from GPS and digital compasses mounted on a camera. Moreover, we have focused on a clustering of geo-tagged photos based on reverse-geocoding with a context such as "walking along a street". We conducted an experiment using 590 geo-tagged photos taken from a GPS camera with a digital compass. Then, we have confirmed that our algorithm can perform geospatial retrieval for a street-view photo browsing with the context.