Abstract:

Among indoor position scenarios,the positioning method based on the Received Signal Strength has several problems such as unstable signal propagation,high computational complexity and low positioning accuracy caused by large target areas.In order to solve these problems,this paper proposes a signal two-scale nearest neighbor with dynamic correction method,according to the connectivity structure of the target area,and the system uses the one-vs-rest support vector machine to construct a partition model of the target area in order to predict the subarea with signal changes.This paper trains the AP signal-distance model based on Gaussian process regression in partition.It is important to realize the correction of signal fluctuation value by predicting the path-loss characteristics in partition.In order to improve the positioning accuracy,by combining signal similarity with signal difference,a two-scale nearest neighbor algorithm is established.The k value of the nearest neighbor is adaptively calculated by combining the environmental parameter in order to reduce the influence of environmental noise.Through simulation experiments,the average localization error of the proposed algorithm is less than 0.5173 m,indicating that the algorithm causes a lower error than the traditional algorithm by more than 25 percent.

Key words: indoor positioning, fingerprint positioning, subarea, Gaussian process regression, signal two-scale