Wi-Fi Indoor Localization based on Multi-Task Deep Learning

Wei-Yuan Lin, Ching-Chun Huang, Nguyen-Tran Duc, Hung-Nguyen Manh

Abstract: Conventional Wi-Fi-based indoor localization methods rely on training a RSS fingerprint model to predict user locations. Most fingerprinting models only consider the distribution of RSS (radio signal strength) at a location and ignore the relationship between adjacent locations. Another challenging issue is the RSS inconsistency problem where the RSSs of neighboring locations are not as similar as the ideal expectation. To address these problems, we suggest well utilizing the richer regional features rather than the raw RSSs. Thereby, we proposed a deep learning network which integrates three components: the One-Dimension-Convolutional Neural Network to extract regional RSS features, the Siamese architecture to handle the similarity inconsistency problem, and the Regression network for user positioning. Our experiments present promising results compared with the state-of-art methods.