Updating Radio Maps Without Pain: An Enhanced Transfer Learning Approach

In recent years, the demand for indoor positioning systems has grown rapidly with regard to location-based services. As a cost-effective choice, WiFi-based indoor positioning has attracted great increasing research attentions because it does not require external devices installed in the target environment. Although extensive research has been conducted on WiFi fingerprint matching, the problem of automatically adapting radio maps to fresh signal space environment still exists. The traditional methods often conduct site surveys regularly to update the outdated radio maps, which is time consuming and laborious. In this work, we propose an indoor positioning system AAIMSS to automatically update the radio maps based on an enhanced transfer learning (TL) approach with altered access points (APs) identification and mapping space searching. In the system, the proposed TL approach removes the outlier features and then searches a more accurate mapping space between the original radio map and crowdsourcing data. Our lightweight solution does not rely on additional devices and inertial sensors with high-power consumption. A set of experiments has been conducted in a teaching building to evaluate AAIMSS. The results show that this AAIMSS is robust to locate users in a dynamic environment. The average positioning accuracy achieves 2.5m, which significantly outperforms the positioning strategies with the original radio map by 65.3%, the radio map by directly removing the altered APs by 19.4%, and the radio map by the traditional TL by 85.2%.