Fuzzy Weighted Localization System based on Transfer Learning

Fingerprint-based indoor positioning system in dynamic environment needs to frequently correct fingerprint database and adjust parameters of positioning model. To alleviate this problem, this paper proposed an adaptive fingerprint positioning system based on fuzzy theory and transfer learning to reduce the cost of fingerprint collection and improve the positioning accuracy. Specifically, the transfer learning algorithm maps source fingerprint features and target fingerprint features into same space using target unlabeled fingerprints and source labeled fingerprints. After that, the fuzzy membership degree is calculated and assigned to different fingerprint positioning models, and then the joint fuzzy weighted model is used to predict the positioning results. The experiment test shows that the proposed system can improve localization accuracy and reduce cost of system.