Research On The Semi-supervised Learning For Indoor Wi Fi Fingerprint Localization System

With the widely use of smart mobile devices and wireless network, the location-based services(LBS) are attracting more and more attention from various sections of society. Demands for LBS are rapidly expanding in fields like public safety, emergency rescue, and goods transportation. In outdoor scenarios, a good performance has been provided by the Global Navigation Satellite Systems(GNSSs), such as Global Positioning System(GPS), Beidou System(BDS), Galileo and other systems. Indoors, however, these satellite navigation systems cannot achieve satisfying accuracy due to signal fading and multipath effect. Therefore, indoor localization has gained increasing interest and become a hot topic in the navigation field these years.Most indoor localization systems are based on signals collected from mobile devices(smart phone, tablet), such as WLAN, Bluetooth, GSM/3G, accelerator, digital compass, camera, etc. Among these WLAN based localization is the most widely applied these years. With broad application of WLAN in the great mass of public places and office areas, few extra devices are needed for WLAN based localization systems.WiFi fingerprint localization system is a typical WLAN based localization system, which contains two phases: offline training phase and online locating phase. In the offline phase, a fingerprint database is built based on RSS vectors at each reference point defined by users. Then in the online phase, current user location is calculated by comparing current RSS vector with the fingerprint database generated beforehand. The offline training phase using a traditional method, however, requires a big amount of reference points and sampling data to guarantee the localization accuracy, which results into a tremendous amount of workload in offline phase. In this article, we propose fast fingerprint training approach:(1) a semi-supervised learning(SSL) is adopted during the training phase. In a target area, users sample the RSS vectors from ‘visible’ WiFi Access Points with a smartphone while walking along the predefined route continuously.(2) Provided with the collected RSS vectors, a Gaussian process method is applied for modeling the RSS distribution of each access point with the target area. Finally, a fingerprint database is generated by merging all the distributions from all the access points deployed in the target area. Given the above approach, the proposed method achieves a fingerprint database effectively and accurately.