Indoor Localization Method Based on WiFi Signals and Building Layout Model

Jure Tuta (2018) Indoor Localization Method Based on WiFi Signals and Building Layout Model. PhD thesis.

Preview

PDF Download (3690Kb)

Abstract

WiFi indoor localization is a difficult task due to the variability of the WiFi signal. Consequently, there have been many attempts to develop WiFi-based methods which were aided by some other means to provide accurate indoor localization. Technologies like dead reckoning and IMU sensors, crowd utilization and pattern matching, specialized Li-Fi hardware and directional antennas, etc. were used to aid the WiFi in order to develop more accurate and stable methods. The main disadvantage of such methods lies in difficult deployments due to technologies and requirements: Dead-reckoning-aided methods are not suitable for stationary objects, methods leveraging groups of people and many individuals are not best suited for home environment, Li-Fi assisted methods require mobile terminals to provide Li-Fi connectivity and therefore rule out mobile phones as the most common terminal. In the past, many fingerprinting methods were proposed; these require a survey in the area of localization during the setup phase. Unfortunately, the majority of fingerprinting-based methods do not address issues of long-term stability of the WiFi signals. Thus, they face accuracy issues a few days after the calibration; frequent, costly and time-consuming recalibration procedures are used to address these issues. Model-based methods try to eliminate calibration procedures by simulating signal propagation. Many of the methods assume at least some parameters of propagation as fixed and therefore poorly address the issues of WiFi’s variability and long-term stability. A pure WiFi model-based method that successfully addresses these issues and requires a mobile terminal only for emitting or receiving the WiFi signals is the ultimate goal of the WiFi indoor localization. This thesis presents a novel indoor localization method, with the main intent of addressing the issues of real-world applicability. Therefore, we focused on developing a method with accuracy comparable to the state-of-the-art methods, while reducing the complexity of deployment and minimizing the required maintenance for long-term deployments. The presented method is a model-based method, implementing self-adaptive operability, i.e. it does not require any human intervention. The thesis discusses in detail the topics of the long-term stability of the WiFi signal, receiving vs. transmitting methods, the future WiFi standards, comparability of the methods and architectural aspects with respect to real-world applicability of the localization methods. Our presented method estimates the parameters of signal propagation, by knowing the positions of the access points, the architectural floor plan with the dividing walls and by monitoring power of the packets travelling between the access points. From this data propagation parameters defined in propagation model are inferred in an online manner. A device trying to define its position captures power information of the packets sent by the access points. Devices’ information on the observed power is used to determine its position by an algorithm run on the localization server. The presented WiFi method is primarily developed and evaluated in single- and multi-room office environments. The method’s ability to be easily applicable in any environment is emphasized by its evaluation in two different environments – office and residential. Between the two, no parameters were modified, thus evaluations indicate universality of the method. Furthermore, we provide evaluation also in narrow hallway because in the field of indoor localization such evaluation environments are common practice. During the evaluation of our proposed method in the office environment, we obtained an average error of 2.63 m and 3.22 m for the single- and multi-room environments respectively. Second evaluation was performed in the residential environment, for which the method or any of the parameters were not modified. Our method achieved an average evaluation error of 2.65 m with standard deviation of 1.51 m, during the four independent evaluations, each consisting of 17 localization points. High accuracy of localization, with acknowledgement to the intricate and realistic multi-room floor plan with different types of walls, realistic furniture and real-world signal interference from the neighboring apartments, proves the method’s applicability to the real-world environment. Evaluation accuracy can be compared to the state-of-the-art methods, while our easily-applicable method requires far less complicated setup procedures and/or hardware requirements. In the second part of the thesis, we generalize the WiFi method to be applicable to the frequencies other than 2.4 GHz WiFi. By defining a fusion algorithm which considers accuracy of the individual frequencies, we have defined the MFAM method: Multiple Frequency Adaptive Model-Based Indoor Localization Method. The MFAM is one of the first purely model-based approaches capable of utilizing multiple frequencies simultaneously. The MFAM method was evaluated in residential environment on two frequency bands: 868 MHz and 2.4 GHz. The method retained positive properties of our WiFi approach (e.g. pure model-based, self-adaptive operability, wide applicability on affordable hardware), while improving the accuracy due to multi-frequency fusion. The usage of multiple frequencies improved the average error of localization from 2.65 m, while using only the WiFi, down to 2.16 m, in the case of multi-frequency fusion, thus improving localization accuracy for 18%. Similar improvements were observed also for the standard deviation. Although the accuracy of the presented WiFi and MFAM methods is comparable if not better than the state-of-the-art methods, one of the most important achievements of our work is the applicability of the method to the real-world situations and its long-term stability. The definition of our method ensures that the accuracy of the method will be the same at the time it is initialized, as well as days later, without any human interaction.

Actions (login required)