RSSI-Based Trajectory Prediction for Intelligent Indoor Localization

Wireless fidelity (Wi-Fi) fingerprint-based localization technique is attracting great attention both from academia and industry due to its ease of deployment and low cost. To achieve high-precision indoor localization, we propose a new hybrid deep neural network (DNN) method based on the received signal strength indicator (RSSI). Compared to traditional algorithms without considering the time correlation, our proposed method takes into account the correlation between each step in the trajectory and utilizes trajectory prediction to assist localization. Specifically, this hybrid DNN uses the stacked auto-encoder (SAE) algorithm for feature reconstruction after data preprocessing, effectively extracting latent codes and reducing feature space. To improve the localization accuracy, the trajectory prediction based on long short-term memory (LSTM) is applied, in which it efficiently establishes the relationship between features and labels using the extracted latent codes, achieving robust and accurate classification. Moreover, a weighted filter is incorporated for further improving localization accuracy. Experimental results show that the proposed RSSI-based trajectory prediction for indoor localization outperforms other baseline schemes and can achieve sub-meter localization accuracy.