基于多高斯混合模型的WLAN室内定位系统

对参考点上信号强度分布采用多高斯混合模型建模从而构建定位系统指纹数据库,模型参数使用期望最大值算法进行估计.由于该模型考虑了各无线接入点信号强度之间的相关关系,因而能更精确地对参考点上信号强度实际分布进行建模,从而使得系统精度提高.测试实验结果表明提出的方法其误差在3m内的概率相对于HORUS,MLP和KCCA分别提高了12%,10%和7%,而平均误差分别降低了0.77m,0.81m和0.20m,验证了本方法的有效性.

WLAN indoor location system based on multi-Gaussian mixture model

The signal strength distribution on the reference was modeled to construct the fingerprint database using multi-Gaussian mixture model,and the parameters of multi-Gaussian mixture model were estimated by using the expectation maximization algorithm.Because this model considers the relationship among signal strengths from different access points,the actual signal strength distribution can be modeled more accurately,and it can greatly improve the system accuracy.Compared with the HORUS,MLP and KCCA,experiment shows that the proposed method can improve the probability of system error which is less than 3 m about 12%,10% and 7% respectively and reduce the mean error about 0.77 m,0.81 m and 0.20 m respectively.This illustrates the effectiveness of the proposed method.