Crowdsourced Indoor Positioning: Integrating 5G NR and WiFi Technologies

Abstract. Indoor positioning technology is a key area of research in location-based services. Crowdsourced WiFi and mobile communication signal fingerprinting are critical for achieving large-scale indoor positioning for consumers. However, existing crowdsourced positioning solutions are not suitable for typical environments like shopping malls due to the need for additional equipment, and their learning methods often have low computational efficiency and generalization ability in complex environments. This paper proposes a system that introduces clustering concepts to repair remaining trajectories using representative trajectories. WiFi SSID and 5G NR SSB data collected along trajectories are used as features for clustering analysis. Reliable starting points are obtained through GNSS accuracy metrics to correct trajectories, and a Bi-LSTM model is utilized to extract trajectory inflection points. Unprocessed trajectories of the same category are corrected based on inflection point features, thereby constructing a WiFi-5G fingerprint database. In addition to providing positioning services, the proposed system iteratively infers the locations of shops, allowing for the construction of a semantic map. The experimental site is the first floor of a large shopping mall, with a dataset comprising 185 user-collected trajectories totaling 2 hours in duration. The trajectory clustering accuracy exceeds 80%, with an average localization error of 5.73 meters for static test points, and an average error of 4.38 meters for the semantic map. Compared to existing crowdsourced solutions, the proposed method shows significant improvements in feasibility, accuracy, and efficiency.