Scalable and Accurate Floor Identification via Crowdsourcing and Deep Learning

Understanding the floor-level location of a user in a multi-storey building is crucial for various applications, including emergency response and shopping guides. Current floor identification systems face several challenges, such as low accuracy, the requirement for time-consuming site surveys, assumptions about user encounters, initial floor knowledge, and poor generalization. In this chapter, we present UnFI, a novel floor identification system that is both scalable and accurate, eliminating the need for site surveys, initial floor knowledge, and other assumptions. The system leverages widely-available smartphone sensors to determine a user's floor location. By automatically recognizing the ground floor and utilizing the stable pressure difference between floors, we avoid the need for cumbersome site surveys for fingerprint association. To ensure precise floor identification, we have developed deep learning-based methods for indoor/outdoor detection and floor identification. Experimental results demonstrate that UnFI outperforms existing systems and shows great potential for large-scale deployment.