Wi-Fi/MARG Integration for Indoor Pedestrian Localization

Zengshan Tian, Yue Jin, Mu Zhou, Zipeng Wu, Ze LiSensors 2016 · DOI

TL;DR

This paper presents a novel indoor pedestrian localization system that fuses Wi-Fi Received Signal Strength (RSS) data with data from on-board Magnetic, Angular Rate, and Gravity (MARG) sensors. By enhancing Pedestrian Dead Reckoning (PDR) and Wi-Fi localization techniques and employing an Extended Kalman Particle Filter (EKPF) for sensor fusion, the system achieves significantly improved accuracy compared to using either technology alone.

What Problem Does This Paper Address?

Accurate indoor localization is a critical challenge, especially for applications like emergency services, where GPS is unreliable. While Wi-Fi Received Signal Strength (RSS) fingerprinting is a popular approach, it suffers from signal fluctuations and large localization errors. Inertial sensors (MARG) offer high accuracy over short periods but accumulate errors over time (drift). This paper addresses the need for a robust, accurate, and low-cost indoor localization system by effectively combining the strengths of both Wi-Fi and MARG sensors to overcome their individual limitations.

Methods

The proposed system integrates several key innovations:

  • Enhanced Pedestrian Dead Reckoning (PDR):
    • Velocity Estimation: Utilizes a Back Propagation (BP) neural network to optimize a stride length model based on accelerometer data. This model is then used in real-time to estimate stride length and velocity.
  • Heading Estimation: Employs a complementary filter auxiliary Extended Kalman Filter (EKF) to fuse data from the accelerometer, gyroscope, and magnetometer. This approach aims to mitigate gyroscope drift and sensor noise for more accurate heading estimation.
  • Enhanced Wi-Fi Localization:
    • Radio Map Construction: Creates multiple radio maps accounting for different human body orientations to improve robustness against signal variations.
  • Affinity Propagation Clustering (APC): Used in the offline phase to group reference points (RPs) and in the online phase for coarse localization, reducing computation time and large localization errors.
  • Weighted K-Nearest Neighbors (WKNN) with Inverse Matching: Applied in the fine localization step to refine the estimated position.
  • Extended Kalman Particle Filter (EKPF) Fusion:
    • This algorithm fuses the enhanced PDR and Wi-Fi localization outputs. The EKPF uses the Extended Kalman Filter (EKF) to optimize the Importance Density Function (IDF) for particle generation, aiming to reduce the number of particles needed and improve the approximation of the posterior distribution. This fusion helps to mitigate PDR drift and Wi-Fi localization jumps.

Datasets

The experiments were conducted in a real indoor environment. The paper does not explicitly mention pre-existing public datasets for training or evaluation. Instead, it describes the setup of their experimental environment, including the layout, the number of Access Points (APs), and Reference Points (RPs).

Key Insights and Ideas

  • Complementarity of Wi-Fi and MARG: The core insight is that Wi-Fi and MARG sensors are highly complementary. Wi-Fi provides absolute position references that can correct PDR drift, while PDR provides continuous motion updates that can bridge gaps in Wi-Fi coverage or accuracy.
  • Addressing Wi-Fi Signal Instability: The paper acknowledges the significant impact of multipath interference and human body obstruction on Wi-Fi RSS. Their approach of creating multiple radio maps for different orientations directly tackles this issue.
  • Mitigating PDR Drift: The use of an EKF-based complementary filter for heading estimation and a BP neural network for stride length estimation are key to improving the accuracy and reducing the cumulative errors inherent in PDR.
  • Efficient Fusion with EKPF: The EKPF is presented as a more efficient fusion method than traditional Particle Filters, achieving better accuracy with fewer particles by optimizing the IDF.

Key Results

  • The proposed Wi-Fi/MARG fusion system achieved an average localization error of 0.85 m.
  • This is a significant improvement compared to using Wi-Fi localization alone (average error not explicitly stated but implied to be higher than 0.85m) or PDR alone (which suffers from cumulative errors).
  • The system demonstrated better performance than conventional PDR-based localization (PBL) and Wi-Fi-based localization (WBL) systems.
  • Compared to other fusion systems like EKF and Robust EKF (REKF), the proposed system achieved a lower mean error (0.85 m vs. 1.21 m for EKF and 1.68 m for REKF) and standard deviation of errors.
  • The heading estimation using the Complementary Filter Auxiliary EKF (CFAEKF) showed superior performance, with most errors falling within 10°, and reduced mean error by over 57% compared to EKF and GDA.

Limitations and Future Work

  • Radio Map Construction Cost: The authors acknowledge that constructing the radio maps, especially with multiple orientations, is time-consuming and labor-intensive.
  • Future Work: The authors plan to focus on developing more effective and efficient radio map construction approaches to reduce the time and labor costs associated with site surveys for large-scale indoor environments.

Summary updated

BibTeX
@article{tian2016wi,
  author = {Zengshan Tian and Yue Jin and Mu Zhou and Zipeng Wu and Ze Li},
  title = {{Wi-Fi/MARG Integration for Indoor Pedestrian Localization}},
  journal = {Sensors},
  year = {2016},
  doi = {10.3390/s16122100},
}
Figure