基于半监督学习的室内定位算法

Indoor Positioning Algorithm Based on Semi-supervised Learning

收集带有位置信息的经验样本即标定样本是一个花费昂贵的工作,限制了基于机器学习方法的实际应用。针对该问题,提出一种基于流形正则化的室内定位算法LocMR,该算法使用少量的标定样本和充足的未标定样本学习得出信号空间到位置空间的映射关系。在实际IEEE 802.11Wi-Fi环境中采集的数据集上进行验证,结果表明,LocMR在达到较高定位精确度的同时,能大幅减少定位系统的工作量,增强了其实际应用能力。

Collecting training data with positioning information is a costly work, which restricts the actual deployment of positioning estimation system and becomes a bottleneck problem. Aiming at rite above problem, this paper presents a positioning estimation algorithm LocMR based on manifold regularization, which is a semi-supervised machine learning algorithm, to learn mapping function with a few labeled data and sufficient unlabeled data. The algorithm LocMR is verified in real IEEE 802.11 Wi-Fi wireless data set, result shows that it reaches higher accuracy, while reduces calibration effort greatly at the same time, thus the application availability, of positioning estimation system is greatly enhanced.