AI-Based Indoor Localization Using RSSI: CNN-LSTM Based Approach

Le résumé fourni par la source

A fundamental challenge of smart environments is indoor localization. However, signals transmitted over Bluetooth Low Energy (BLE) are sensitive to noise, multipath effects, and attenuation. Classical fingerprinting methods, such as K-Nearest Neighbors (KNN), Support Vector Machines (SVM), and Random Forest (RF), typically achieve less than 70% accuracy in dynamic environments, limiting their practical use. This paper proposes a hybrid deep learning model that incorporates a denoising autoencoder within an ensemble prediction pipeline consisting of a denoising autoencoder, convolutional neural networks (CNN), and long short-term memory (LSTM) units. The proposed CNN-LSTM model achieves an accuracy of 85.5% on a public BLE RSSI dataset containing 105 locations in a single run, and improves to 87.7% accuracy using stratified cross-validation ensembles. Further analysis of the confusion matrix and error distribution reveals robust performance in beacon-rich regions, while error tails correlate with multipath effects and beacon-sparse areas. This approach demonstrates the potential to achieve scalable accuracy, robustness, and computational efficiency, making enhanced hybrid deep learning models a promising standard for viable indoor positioning systems.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.