Crowdsourced 3-D Wi-Fi AP Localization in Multifloor Buildings via Vertical Transitions

Range-based positioning systems require accurate locations of anchor nodes; however, obtaining these locations is often costly and time-consuming. In this paper, we propose a novel framework to automatically estimate the locations of Wi-Fi access points (APs) using unlabeled data passively collected from mobile users as they naturally navigate the environment. The proposed framework is primarily based on the observation that vertical movement within a building is constrained to a limited number of transition points, such as escalators and elevators. Accordingly, the framework first detects vertical transitions of users by analyzing barometer readings. By simultaneously considering the transitions of all users, it determines the absolute floor level of each user throughout the building. Next, user trajectories are reconstructed using inertial sensor measurements and partitioned into segments corresponding to individual floors. A cost function is designed to align each segmented trajectory on the determined floor level by selecting the most probable entry and exit points where vertical transitions are possible. The locations of APs are then estimated using Wi-Fi signal strengths measured along the aligned trajectories. The effectiveness of the proposed framework was evaluated in a large-scale, 10-floor shopping mall, where each floor spans up to 30,000 m. Using unlabeled datasets collected from multiple users, the locations of 1,214 APs operating on the 2.4 GHz band were estimated. Based on this database, user locations could be obtained during the online phase, demonstrating an average accuracy of less than 5 m. This result confirms the successful operation and practical applicability of the proposed framework.