Enhancing Indoor User Localization: An Adaptive Bayesian Approach for Multi-Floor Environments

Indoor localization systems are crucial in addressing the limitations of traditional global positioning system (GPS) in indoor environments due to signal attenuation issues. As complex indoor spaces become more sophisticated, indoor localization systems become essential for improving user experience, safety, and operational efficiency. Indoor localization methods based on Wi-Fi fingerprints require a high-density location fingerprint database, but this can increase the computational burden in the online phase. Bayesian networks, which integrate prior knowledge or domain expertise, are an effective solution for accurately determining indoor user locations. These networks use probabilistic reasoning to model relationships among various localization parameters for indoor environments that are challenging to navigate. This article proposes an adaptive Bayesian model for multi-floor environments based on fingerprinting techniques to minimize errors in estimating user location. The proposed system is an off-the-shelf solution that uses existing Wi-Fi infrastructures to estimate user’s location. It operates in both online and offline phases. In the offline phase, a mobile device with Wi-Fi capability collects radio signals, while in the online phase, generating samples using Gibbs sampling based on the proposed Bayesian model and radio map to predict user’s location. Experimental results unequivocally showcase the superior performance of the proposed model when compared to other existing models and methods. The proposed model achieved an impressive lower average localization error, surpassing the accuracy of competing approaches. Notably, this noteworthy achievement was attained with minimal reliance on reference points, underscoring the efficiency and efficacy of the proposed model in accurately estimating user locations in indoor environments.