Wifi Indoor Positioning with Genetic and Machine Learning Autonomous War-Driving Scheme

Pham Doan Tinh,Bui Huy Hoang

doi:10.14569/ijacsa.2022.0130279

Abstract

Wifi Fingerprinting is a widely used method for indoor positioning due to its proven accuracy. However, the offline phase of the method requires collecting a large quantity of data which costs a lot of time and effort. Furthermore, interior changes in the environment can have impact on system accuracy. This paper addresses the issue by proposing a new data collecting procedure in the offline phase that only needs to collect some data points (Wi-fi reference point). To have a sufficient amount of data for the offline phase, we proposed a genetic algorithm and machine learning model to generate labeled data from unlabeled user data. The experiment was carried out using real Wi-fi data collected from our testing site and the simulated motion data. Results have shown that using the proposed method and only 8 Wi-fi reference points, labeled data can be generated from user’s live data with a positioning error of 1.23 meters in the worst case when motion error is 30%. In the online phase, we achieved a positioning error of 1.89 meters when using the Support Vector Machine model at 30% motion error.

Highlights

As people spend more time indoors, many location-based applications and services require to known user indoor location
While the global positioning system (GPS) is a popular positioning method, it can hardly be applied to indoor environments because lacking in line of sight (LOS)
The approach assumed that the Receive Signal Strength Indicator (RSSI) measurement from access points (APs) for every location is unique

Summary

INTRODUCTION

As people spend more time indoors, many location-based applications and services require to known user indoor location. In the offline phase of conventional Wi-fi Fingerprinting, the coordinate system of the whole area needs to be built, numerous data points are marked to have their RSSI measurements collected. This procedure consumes a lot of time and effort. The proposed method is different from others because the user motion data are not used together with Wi-fi to directly predict the user location but they are only utilized in the offline phase to generate more labeled data for the machine learning model.

THE PROPOSED SCHEME

Implementation of Genetic Algorithm and Machine Learning Block

Fitness Function Evaluation

CONCLUSION