摘要： 基于无线信号的室内定位技术所采用的定位算法在很大程度上受到无线信号中噪声数据的影响,性能得不到保证。为此,提出一种受约束的k最邻近(kNN)回归算法,提升原始kNN算法对噪声的鲁棒性。假设噪声点对于预测不同测试样本的影响各不相同,通过训练集(含噪声)划 分,即以数据驱动的方式把噪声点划分到合适的子集中,并且限定测试样例的最优近邻搜索空间为其最近邻子集的方式来约束噪声对于kNN算法的影响。实验结果表明,在蓝牙指纹数据的室内定位中,受约束的kNN回归算法明显优于对比算法,达到2.4 m的定位精度,基本满足室 内定位的应用要求。

关键词: 无线蓝牙, 室内定位, 受约束的k最邻近, 信号强度, 噪声数据

Abstract: Wireless indoor positioning techniques are greatly affected by noise data and the performance can not be guaranteed when using these location algorithms.This paper proposes a Constrained k-Nearest Neighbor(CkNN) regression for bluetooth -based indoor positioning which significantly improves the effectiveness and robustness of original kNN.It argues that the affect of the noise in the training set differs from different test instance.If the training set can be partitioned into several disjoint subsets where the noisy points are appropriately assigned,and constrain the search space of nearest neighbors of the test point into its nearest subset,the impact of noisy points in the training set can be minimized.Experimental result shows that CkNN regression method outperforms other algorithms by a large margin,achieving the accuracy of 2.4 m which satisfies the requirement of many position based services.

Key words: wireless bluetooth, indoor positioning, Constrained k-Nearest Neighbor(CkNN), signal strength, noise data

中图分类号: 

• TP181