Local Batch Normalization-Aided CNN Model for RSSI-Based Fingerprint Indoor Positioning

Abstract

Indoor positioning systems have become increasingly important due to the limitations of GPS in indoor environments, such as non-line-of-sight conditions and weak signal strength. Among the various indoor positioning techniques, fingerprinting-based approaches utilizing WiFi signals are highly regarded for their accessibility and convenience. However, existing convolutional neural network (CNN) models for fingerprinting often struggle to maintain consistent performance under diverse environmental conditions. To address these challenges, this study proposes a local batch normalization (LBN)-aided CNN model for received signal strength indicator (RSSI)-based indoor positioning. The LBN technique is designed to overcome the limitations of traditional batch normalization (BN) and layer normalization (LN) in managing location-dependent RSSI variations, thereby improving positioning accuracy. The proposed approach consists of two phases: an offline phase, where RSSI data are collected at reference points to train the model, and an online phase, where real-time RSSI data are used to estimate the device's location. Experimental results demonstrate that the proposed LBN-aided CNN model achieves an accuracy of 92.9%, outperforming existing CNN-based methods. These findings confirm the effectiveness of LBN in enhancing CNN performance for indoor positioning, particularly in challenging environments with significant signal variability.