Crowdsourcing Indoor Positioning by Light-Weight Automatic Fingerprint Updating via Ensemble Learning

In recent years, Wi-Fi-based indoor positioning has attracted increasing research attention due to its ubiquitous deployment. Although extensive research has been conducted on Wi-Fi fingerprint-based positioning, especially, in complex environments and long-term deployments, the automatic adaptation of radio map has not been fully studied and the problems remain open. When the positions of some Access Points (APs) change, the traditional approach regularly conducts site surveying which is time-consuming and labor-intensive. In this paper, we propose a crowdsourcing indoor positioning approach based on ensemble learning for automatic Altered APs Identification and Fingerprints Updating, namely AAIFU. We propose an algorithm to detect and identify the altered APs in crowdsourcing data. After getting the altered APs, we use the relationship between the received signal strength values of the altered APs and the unaltered APs in the crowdsourcing data to train a prediction model used to update the radio map. We also handle the device diversity on all the processes of AAIFU. Our proposed solution is light-weight which does not rely on additional infrastructure and inertial sensors with high power consumption. The comprehensive experiments have been carried out in our teaching building to evaluate the effectiveness of AAIFU. The results show that our proposed AAIFU can effectively adapt the radio map to the movement of APs and improve positioning accuracy. Correspondingly, we achieve an average positioning accuracy of $2.6m$ which outperforms the fingerprinting approach with the original radio map by 63.9%.