A Fast and Precise Indoor Localization Algorithm Based on an Online Sequential Extreme Learning Machine

Han Zou, Xiaoxuan Lu, Hao Jiang, Lihua Xie

TL;DR

This paper introduces an indoor localization algorithm using an Online Sequential Extreme Learning Machine (OS-ELM) to overcome the high costs and inflexibility of traditional WiFi-based fingerprinting methods. OS-ELM offers fast learning and continuous adaptation to environmental changes, leading to improved accuracy and reduced maintenance.

What Problem Does This Paper Address?

Indoor positioning systems (IPSs) are crucial for location-based services, but existing WiFi-based solutions face significant challenges. The dominant fingerprinting approach requires extensive offline "site surveys" to map WiFi signal strengths, which is time-consuming and labor-intensive. Furthermore, these systems are often brittle, failing to adapt to dynamic indoor environments (e.g., changes in occupancy, doors opening/closing) without costly recalibration. This leads to a trade-off between accuracy and the ability to maintain performance in real-world, ever-changing conditions.

Methods

The core of this work is the application of an Online Sequential Extreme Learning Machine (OS-ELM) for indoor localization. OS-ELM is a type of neural network that offers:

  • Fast Learning: Inheriting from Extreme Learning Machines (ELMs), it achieves rapid training speeds.
  • Online Sequential Learning: Unlike traditional batch learning, OS-ELM can learn from data incrementally, one sample or a "chunk" at a time, without needing to retrain on the entire dataset. This is key for adapting to environmental changes.
  • Flexibility: It can handle various hidden node types and learn data in chunks of varying sizes.

The proposed algorithm has three main phases:

  1. Offline Calibration Phase: A sparse WiFi RSS "radio map" is collected at known locations. This data is used to train an initial OS-ELM model. This phase is significantly less intensive than traditional methods due to OS-ELM's efficiency.
  2. Online Sequential Learning Phase: As the environment changes or new data becomes available, OS-ELM can update its model incrementally using newly collected WiFi RSS fingerprints from online calibration points. This allows the system to adapt to dynamics without full recalibration.
  3. Online Localization Phase: When a user requests their location, their current WiFi RSS fingerprint is fed into the latest OS-ELM model to estimate their position.

The paper also details the selection of key OS-ELM parameters, including the hardlim activation function and an optimal number of 950 hidden nodes, determined through cross-validation and repeatability analysis.

Datasets

The paper describes both simulated and real-world experimental data:

  • Simulated Data: A 20m x 20m room was simulated with four WiFi access points. WiFi RSS fingerprints were generated using the ITU indoor propagation model. This simulation involved collecting data at offline calibration points, online calibration points, and online testing points under various scenarios.
  • Real-World Data: Experiments were conducted in the Internet of Things Laboratory at Nanyang Technological University (approx. 580 m²). Data was collected using a Samsung I929 Galaxy SII mobile phone running a custom Android app. This involved:
  • Offline Phase: 30 offline calibration points with 1000 WiFi RSS fingerprints collected at each.
  • Online Phase: Data collected over five days, with two distinct online calibration points and two distinct online testing points selected daily to reflect environmental dynamics.

Key Insights and Ideas

  • Adaptability is Key: The most significant insight is that indoor localization systems must adapt to environmental changes to maintain accuracy. OS-ELM's online sequential learning capability directly addresses this, making it inherently more robust than static fingerprinting databases.
  • Reduced Offline Burden: OS-ELM drastically cuts down the time and effort required for the initial site survey. A "sparse radio map" is sufficient for initial training, which is a major practical advantage.
  • Efficiency in Updates: The online sequential learning is not just adaptive but also very fast. Updating the model with thousands of new data points takes mere seconds, making real-time adaptation feasible.
  • Hardlim Function Performance: The hardlim activation function was found to be superior for this task compared to RBF and sine functions, offering a good balance of accuracy and stability.

Key Results

  • Improved Accuracy: The latest updated OS-ELM model achieved an average localization accuracy of 1.794 m in simulations, a significant improvement over the initial OS-ELM (3.103 m) and enhancing precision by 42.18%.
  • Fast Online Updates: Calculating output weights for 1000 newly received WiFi RSS fingerprints took only 0.014 s on average in simulations.
  • Real-World Performance: In experiments, OS-ELM achieved an average accuracy of 1.973 m when considering online updates from 38 calibration points. This was a 36.31% improvement over KNN, 27.68% over fuzzy KNN, and 23.56% over batch ELM.
  • Robustness to Dynamics:
    • Human Presence: OS-ELM's accuracy remained nearly constant (2.183 m vs. 2.197 m) under human interference, while batch ELM degraded significantly (2.203 m vs. 3.106 m).
  • Door Status: OS-ELM maintained accuracy (2.107 m vs. 2.085 m) whether doors were open or closed, whereas batch ELM's accuracy dropped from 2.214 m to 4.203 m.

Limitations and Future Work

  • Lab Environment Focus: The current experiments were primarily conducted in a lab environment. The authors suggest extending experiments to other indoor scenarios like lecture halls or shopping centers for more conclusive evaluations.
  • Device Heterogeneity: The study used a specific mobile device. The authors note that different devices with varying WiFi chipsets can lead to degraded accuracy. Future work should explore how to improve OS-ELM to handle this heterogeneity.
  • Other Environmental Factors: While occupancy and door status were tested, other environmental factors could also impact performance and warrant further investigation.

Summary updated

Abstract

Nowadays, developing indoor positioning systems (IPSs) has become an attractive research topic due to the increasing demands on location-based service (LBS) in indoor environments. WiFi technology has been studied and explored to provide indoor positioning service for years in view of the wide deployment and availability of existing WiFi infrastructures in indoor environments. A large body of WiFi-based IPSs adopt fingerprinting approaches for localization. However, these IPSs suffer from two major problems: the intensive costs of manpower and time for offline site survey and the inflexibility to environmental dynamics. In this paper, we propose an indoor localization algorithm based on an online sequential extreme learning machine (OS-ELM) to address the above problems accordingly. The fast learning speed of OS-ELM can reduce the time and manpower costs for the offline site survey. Meanwhile, its online sequential learning ability enables the proposed localization algorithm to adapt in a timely manner to environmental dynamics. Experiments under specific environmental changes, such as variations of occupancy distribution and events of opening or closing of doors, are conducted to evaluate the performance of OS-ELM. The simulation and experimental results show that the proposed localization algorithm can provide higher localization accuracy than traditional approaches, due to its fast adaptation to various environmental dynamics.
BibTeX
@inproceedings{zou2015fast,
  author = {Han Zou and Xiaoxuan Lu and Hao Jiang and Lihua Xie},
  title = {{A Fast and Precise Indoor Localization Algorithm Based on an Online Sequential Extreme Learning Machine}},
  booktitle = {Sensors},
  year = {2015},
  url = {https://openreview.net/forum?id=_vzRuGKlXM5},
}

Loading related content...

Figure