Semi-supervised Learning Based Indoor WLAN Positioning Algorithm Using Support Vector Regression

With the rapid development of communication technology and the popularity of intelligent mobile terminals, the demand for location-based services is increasing. In the field of navigation, emergency rescue, and other personalized communication applications, Location-based services play a crucial role. The satellite-based positioning technology such as GPS of the United States, and cellular network-based positioning technology are the two of the most representative outdoor localization technology. However, due to the complexity of the indoor environment, outdoor positioning technology is difficult to satisfy the positioning accuracy at indoor scenario. Meanwhile, with the wide deployment of WLAN infrastructures in most indoor environments, we can locate the target with the help of WLAN in a cost-efficient manner. Since WLAN-based indoor positioning technology is with high positioning accuracy and has no additional requirement for equipment, it has been significantly studied in the past decades. The location fingerprinting-based positioning technology is mostly researched in WLAN positioning technology, which is divided into online and offline phases. And this paper focuses on location fingerprint-based localization algorithm.First, Due to the complex indoor environments and propagation characteristics of the wireless signal, the received signal strength has significant uncertainty and non-linear characteristics. And thus it has a great impact on the performance of location fingerprint-based WLAN indoor localization. At the same time, SVM has a good generalization ability to solve nonlinear problems as well as the cases with small amount of samples. On the basis, the support vector regression is introduced into the indoor WLAN positioning fingerprint to build a mapping prediction model between RSS and corresponding physical locations for the sake of improving positioning accuracy. Secondly, the construction of fingerprint usually takes a lot of cost in offline phase while the mobile terminal can easily obtain unlabeled RSS. Semi-supervised learning can make good use of location-independent RSS which not only reduces the amount of fingerprints, but also improves the positioning accuracy. Therefore, this paper proposes a semi-supervised learning based indoor WLAN positioning algorithm using support vector regression to improve positioning accuracy, which combines semi-supervised co-training algorithm and support vector regression algorithms. Finally, the modified semi-supervised learning based indoor WLAN positioning algorithm using support vector regression is proposed to improve the performance.In this paper, the performance of proposed algorithm is verified in the simulation environment as well as the real indoor environments including office and corridors. By comparing with the traditional algorithms, it is proved that the proposed localization algorithm has advantage with respect to the positioning performance.