基于CSI和加权混合回归的室内定位方法

Indoor Localization Method Based on CSI and Weighted Mixed Regression

近年来,Wi-Fi感知凭借低成本、非接触、不受光照影响、隐私性好等优势,成为人机交互的新兴研究方向。为了提高室内定位技术的精度,提出了一种基于信道状态信息(Channel State Information,CSI)的加权混合回归(Weighted Mixed Regression,WMR)室内定位算法WMR_SKR。该方法分为离线训练和在线预测两个阶段。离线阶段单独训练支持向量回归(Support Vector Regression,SVR)和K近邻回归(K-Nearest Neighbor Regression,KNR)模型,并获得最优的权重分配,建立加权混合回归模型WMR_SKR。在线阶段通过WMR_SKR模型实时预测目标的位置。实验结果表明,本文的WMR_SKR模型在视距环境中82%的概率下定位精度能够达到1 m,非视距环境中80.6%的概率下达到1.5 m,且平均误差和标准误差均小于1.5 m。WMR_SKR模型有效融合了SVR和KNR的优点,提高了室内定位技术的性能。

In recent years,Wi-Fi sensing with the advantages of low cost,non-contact,not being affected by light,and good privacy has become a new research domain of human-computer interaction.To improve the accuracy of the indoor localization method in the real environment,a weighted mixed regression indoor localization method(WMR_SKR)based on channel state information(CSI)is proposed.The localization process is divided into two phases:off-line and on-line.In the off-line phase,the support vector regression(SVR)and K-nearest neighbor regression(KNR)models are trained independently and the optimal weights are obtained.In the on-line phase,the target position is predicted by the WMR_SKR model.The experimental results show that the WMR_SKR can reach 1 m with 82%probability in light of sight environment,1.5 m with 80.6%probability in non-line of sight environment,and the average error and standard error are less than 1.5 m.The WMR_SKR model integrates the advantages of SVR and KNR effectively and improves indoor localization technology’s performance.