부산대학교 도서관

WiFi fingerprinting is a widely used technique for indoor localization, leveraging existing infrastructure to estimate a user's location based on received signal strength (RSS) measurements. The popular used WiFi fingerprinting is K-nearest Neighbors (KNN), which assumes that there is a linear relationship between the WiFi signal distance and real space distance. However, such assumption often fails to hold in complex indoor environments, resulting in significant positioning errors of KNN methods. In this paper, we propose DeepFML, a novel deep learning-based approach for KNN-based fingerprinting positioning, which learns a mapping function to transform raw RSS measurements into features, improving the consistency between the spatial distance in location space and the distance in feature space. Our experiments in complex indoor environments show that DeepFML outperforms state-of-the-art methods, improving the positioning accuracy by about 10% compared to the popular KNN method.