SemSense: Automatic construction of semantic indoor floorplans

Availability of semantic-rich indoor floorplans; where places are labeled with their business names or categories; enables ubiquitous deployment of a wide range of indoor location-based services. In this paper, we present SemSense: a crowdsourcing-based system for automatic enrichment of indoor floorplans with semantic labels. SemSense exploits phone sensors data collected from users during their normal check-ins to location-based social networks (LBSNs) and combines them with data extracted from the LBSNs databases to associate a venue name with its location on an unlabeled floorplan. At the core of SemSense are different modules for handling incorrect location estimates, fake check-ins, as well as increasing the coverage of indoor venues by means of a novel category inference technique. Our experimental evaluation of SemSense using different Android phones in four malls in two cities shows that it can achieve a high semantic labeling accuracy of 87% using a relatively small number of check-ins at each venue in the presence of up to 50% erroneous check-ins. In addition, the proposed coverage extension technique leads to more than 27% enhancement in the places coverage ratio compared to the current LBSNs.