An AP placement strategy for fingerprint based indoor localization

Abstract

Localization in indoor areas deals with the serious problem of collecting data over a broad experimental region while maintaining the location points. The proposed work focuses on addressing the problem of finding the optimal placement of Access Points (APs) to improve AP coverage for a specific set of location points, while minimizing the number of APs required. The objective is to achieve accurate localization with minimal error, ideally less than 2.5 meters. The problem is formulated as a graph coloring problem, where each location point is represented as a node in the graph, and the goal is to assign APs (colors) to the nodes in a way that minimizes interference and maximizes localization ability. To solve this problem, the proposed algorithm, called GLOC_coloring, is introduced. The algorithm begins by applying a machine learning classifier to obtain a confusion matrix, which is used to create a weighted graph. Each vertex in the graph represents a unique location, and edges are added between vertices that have confusion in distinguishing them based on Received Signal Strength Indicator (RSSI) values. The weights of the edges are determined by the confusion matrix. The algorithm proceeds by assigning weights to each vertex based on the sum of weights of all edges connected to it. Starting with the vertex having the maximum weight, a color is assigned to it, and the process continues by selecting vertices with maximum weights from the set of connected vertices. Colors are assigned to these vertices in such a way that they differ from the colors assigned to their connected vertices. This process continues until all vertices are assigned a color, and the number of colors used represents the minimum number of APs required for improved localization accuracy. The experiment took place in two rooms of a benchmark indoor localization dataset. For one single location point up to seven confusing (in terms of location identification) neighbour location points were obtained. This challenge was addressed with proposed GLOC-algorithm and observed that 3 new Aps are sufficient to reduce the localization confusion and their positions were found on the available floor map of the dataset.