An adaptive Wi-Fi indoor localisation scheme using deep learning

Indoor localisation is an important issue for many indoor applications. Many deep learning-based indoor localisation schemes have been proposed. However, these existing schemes cannot adjust according to different environment. To improve the existing schemes, a novel indoor localisation scheme, which can adaptively adopt the proper fingerprint database according to the collected signals, is proposed in this paper. The proposed Wi-Fi indoor localisation scheme uses two fine-tuning algorithms, namely the cross entropy and the mean squared algorithms, to build the corresponding fingerprint databases. When the standard deviation of the collected signals does not exceed the threshold, the fingerprint database built by the cross entropy algorithm is adopted; when the standard deviation of the collected signals exceed the threshold, the fingerprint database built by the mean squared algorithm is adopted. The experimental results show that the proposed scheme can improve the accuracy of the training data and reduce the localisation error.