An Efficient Method for BLE Indoor Localization Using Signal Fingerprint

The rise of Bluetooth Low Energy (BLE) technology has opened new possibilities for indoor localization systems. However, extracting fingerprint features from the Received Signal Strength Indicator (RSSI) of BLE signals often encounters challenges due to significant errors and fluctuations. This research proposes an approach that integrates signal filtering and deep learning techniques to improve accuracy and stability. A Kalman filter is employed to smooth the RSSI values, while Autoencoder and Convolutional Autoencoder models are utilized to extract distinctive fingerprint features. The system compares random test points with a reference database using normalized cross-correlation. Performance is assessed based on metrics such as the number of reference points with the highest cross-correlation (), average localization error, and other statistical indicators. Experimental results show that the combination of the Kalman filter with the Convolutional Autoencoder model achieves an average error of 0.98 meters with . These findings indicate that this approach effectively reduces signal noise and enhances localization accuracy in indoor environments.