基于全局流形结构的半监督学习室内定位算法

Indoor Localization Algorithm Based on Semi-supervised Learning of Global Manifold Geometry

针对传统基于WLAN的指纹定位方法中,因需要密集采集参考点标记数据带来的指纹库构建工作开销大的问题,提出一种基于全局特征保持的半监督流形对齐指纹库构建算法。该算法仅利用少量采集时间开销大的标记数据,结合大量易采集的未标记数据,通过求解流形对齐目标函数,实现对未标记数据的位置标定,减少指纹库构建的消耗。同时,运用测地线距离得到全局流形结构以充分挖掘少量标记数据中的对应特征,提升在少量标记数据情况下构建的指纹库精度。真实场景的实验结果表明,所提算法可以显著降低离线指纹库的构建开销,同时可以取得较优的定位精度。

The construction of radio map is time consuming and labor intensive in the conventional wireless local area network (WLAN) indoor localization systems.In order to solve this problem, the paper proposes a semi-supervised manifold alignment radio map construction approach based on the global geometry of manifold structure.The proposed method utilizes a small number of labeled RSS which requires a huge time consuming to collect and plenty of unlabeled data that is easy to obtain.Then, the locations of plenty of unlabeled data can be obtained by calibrating the solution of the manifold alignment of objective function.In addition, the geodesic distance is utilized to capture the global geometry of manifold feature which can fully exploit the correspondence characteristics of the labeled RSS and its coordinates.Thus, it can improve the accuracy of radio map with limited labeled RSS data.The extensive experiments demonstrate that the proposed method can construct an accurate radio map at a low manual cost, as well as achieve a high localization accuracy.