基于半监督仿射传播聚类和KLDA的室内定位算法*

室内定位中位置指纹库采集的密集程度往往跟定位精度密切相关,针对离线阶段时指纹库稀疏的情况下定位精度低的问题,提出了一种基于半监督仿射传播聚类和KLDA的室内定位算法。该算法结合了在线阶段采集无位置标签的RSSI数据,通过建立局部邻域图将无位置标签的RSSI信息反映到离线指纹数据的结构中,并使用KLDA方法抽取位置指纹库中最大的特征信息,有效利用了无位置标签的RSSI信息从而提高定位精度。实验结果表明,该算法结合在线阶段RSSI数据后定位精度得到了明显的提高。而且在仅保留离线指纹数据库三分之二的情况下,也几乎能够取得与传统KNN算法使用全指纹库时相同的定位精度,相当于减少了离线阶段采集指纹库的工作开销。

Indoor Localization Algorithm Based on Semi-supervised Affinity Propagation Clustering and KLDA

Positioning accuracy is often closely related with the intensity of location fingerprint for indoor localization. When location fingerprint collected during offline stage is sparse, the accuracy is very low. Aiming at solving the problem above, an indoor localization algorithm based on semi-supervised affinity propagation clustering and KLDA is put forward. The algorithm takes advantages of unlabeled data and reflects information of it into the structure of offline fingerprint through establishing the local neighborhood graph. It uses KLDA method to extract the greatest feature of the location fingerprint and effectively uses unlabeled data, thus to improve positioning accuracy. The experimental results show that the algorithm combining with unlabeled data collected online can improve the localization accuracy obviously. In addition, when keeping only two-thirds of offline fingerprint database, it also can achieve almost the same positioning accuracy as the traditional KNN algorithm using complete database, which is equivalent to reduce working overhead of collecting fingerprint in the offline phase.