Quick Overview of Indoor Localization

The localization requirements in different building layouts are significantly different. The techniques suitable for each case vary accordingly. Here list two indoor localization scenarios,

• an open space (single-zone), such as stadium, swimming pool, fitness center, lecture theater, airport waiting zone, supermarket, etc. In such an open space, no major obstacles in the middle of a space, so that either wireless signal or light emits in a line-of-sight (LOS) fashion, which means a signal sink is able to receive signal directly from a signal source.
• a multi-floor and multi-room building, such as apartment, hospital, college hall, etc. Such a building usually is subdivided into multiple compartments, rooms, units by visible bearing walls. These bearing walls and well-designed obstacles hinder the LOS signal. The main challenge to the techniques based on location fingerprints is that the received signal strength could be affected by diffraction, reflection, scattering, and absorption during the propagation in indoor environments.

For instance, if our objective is to localize a person in a standard supermarket, it can be categorized into localization in an open space.

The localization precision is mainly supportive to reach one of the three levels:

• a section, such as deli, fruits, meats, soaps, dairy, freezer, etc.
• an aisle, such as A6, A8, etc.
• specific within a spot with a +- 1m deviation. So far in academia, the average minimum deviation of +- 0.4m is acceptable, while some researches achieve +- 0.3m deviation.

Methodology

Primarily, the state-of-art techniques consist of

1. wireless signal: 2.4GHz, 5.8GHz. The applicable localization principles basically consist of fingerprinting, trilateration, and triangulation. In most practical experiments, fingerprinting gains the highest acceptance.

• WiFi-based In an existing environment, the quantity of WiFi access points (APs) should be adequate to distinguish the desired locating resolution. Typically, no particular hardware needs to apply on a handheld client; software development is needed for sure. At least, one computing localization server is needed.

• Bluetooth-based IBeacons are needed to well-deployed in the site. Typically, no particular hardware needs to apply on a handheld client; software development is needed for sure. At least, one computing localization server is needed.

2. Visible Light Communication (VLC) One feature of LED is that it can be turned on/off instantaneously, which makes it possible to communicate over visible light. Off-the-shelf LED bulbs are capable of transmitting signals embedded in the visible light, such that we leverage the illumination system for accurate localization. Each LED light source beacons information including ID and location on a certain optical channel. The client (smartphone) receives and decodes the signal with a light sensor. The distance from the client to each light source is calculated with a model based on the received signal strength (RSS), and the location is determined using trilateration.

Major methods include

• Optical Angle-of-Arrival (AOA) Localization
• Optical Time-of-Arrival (TOA) Localization
• Fingerprinting of RSS
• Model-based LED luminaire needs device configuration for its particular frequency of emitting code. Conventional receiver types can be either camera or photo-diode.

3. Computer Vision-Based Detecting and tracking somebody by using images captured. This technique is costly because it requires high-resolution cameras, face/body identification data storage, and heavy computations.

4. Magnetic Interference In some researches, magnetic field and even the value of GPS signal indoors (its level) has been used as fingerprints. Oftentimes, this technique can be leveraged as an assistive feature working with WiFi-based approach.

Principle of Fingerprinting Algorithm

The scene analysis algorithm refers to the type of algorithm that first collects features (fingerprints) of a scene and then estimates the target place of an object by matching online measurements with the closest ‘a priori’ location fingerprints. RSS-based location fingerprinting algorithm is commonly used in scene analysis.

The traditional fingerprinting approach can be divided into two stages: offline stage and online stage. During the offline stage, a site survey is performed in an environment. The location coordinates and respective signal strengths from nearby WiFi access points are collected. During the online stage, a location positioning technique uses the currently observed signal strengths and previously collected information to figure out an estimated location. The main challenge to the techniques based on location fingerprints is that the received signal strength could be affected by diffraction, reflection, scattering, and absorption during the propagation in indoor environments.

The classification process can be template matching, nearest-neighbor by Euclidean distance, machine learning, deep learning, etc. Note that in an open space, feeding all AP RSSIs into machine learning training calls for high computation but with bad accuracy unfortunately.

Here is one paper recommended

X. Wang and Y. Feng, “An Ensemble Learning Algorithm for Indoor Localization,” 2018 IEEE 4th International Conference on Computer and Communications (ICCC), Chengdu, China, 2018, pp. 774–778.

Abstract: In this paper, we investigated the problem of localizing a smartphone with iBeacon signal strengths utilizing an ensemble learning algorithm. We built a real testing environment and examined the performance of the ensemble learning algorithm in our positioning system that outperformed any single classifier individually. We also proposed two approaches to improve the accuracy: Exponentially Weighted Moving Averages (EWMA) to deal with wireless signal fluctuation, and data augmentation for enlarging existing data volume. Further, the extent of the density received signal affecting the accuracy of localization by using different intervals was discussed. Most importantly, we tested our algorithm in a real environment. In order to combat overfitting, data balancing on training datasets in each reference point was introduced. By a series of comprehensive experiments, we have corroborated that the weighted fusion algorithm is capable of localization with high accuracy.

URL: http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=8780770&isnumber=8780575