RadarPDR: Radar-Assisted Indoor Pedestrian Dead Reckoning

Jianbiao He, Wei Xiang, Qing Zhang, Bang WangSensors 2023 · DOI

TL;DR

This paper introduces RadarPDR, a novel system that uses an FMCW radar to assist smartphone-based pedestrian dead reckoning (PDR) for more accurate indoor tracking. By calibrating radar-measured wall distances and fusing them with inertial sensor data using a particle filter and Kalman filter, RadarPDR significantly reduces accumulated tracking errors compared to traditional PDR methods.

What Problem Does This Paper Address?

Indoor pedestrian tracking is crucial for many applications, but GPS signals are unavailable indoors. Pedestrian Dead Reckoning (PDR) is a common solution, relying on inertial sensors (accelerometer, gyroscope) in smartphones to estimate movement based on steps and direction. However, these sensors suffer from drift and measurement errors, leading to accumulating inaccuracies over time. Existing methods try to mitigate this by using latent information (like building layouts) or auxiliary sensors (like Wi-Fi, Bluetooth). While effective, these often require prior knowledge of the environment or expensive additional hardware. This paper tackles the challenge of improving PDR accuracy without relying on pre-existing maps or complex infrastructure, by leveraging a readily available, albeit noisy, sensor: an FMCW radar.

Methods

The core of RadarPDR is the integration of an FMCW radar with inertial sensors. The system comprises several key modules:

  • Data Sensing and Preprocessing: Raw data from accelerometers, gyroscopes, and the radar are collected and segmented into step-wise data.
  • Inertial PDR: Standard algorithms are used to estimate step length and heading increments from inertial sensor data.
  • Wall Model Calibration: This is a crucial novel component. It addresses two main issues with radar measurements:
  • Corner Detection and Operation Indicator: An indicator is assigned to each step to determine if it's in a "corner zone" (requiring suspension or reset of operations) or a "flat-segment zone" (allowing for operation or reset). This adaptive approach prevents erroneous adjustments.
  • Wall Distance Correction: A two-stage strategy is employed to handle noise in radar distance measurements, which can be caused by irregular indoor layouts (barriers, corners). This involves a Kalman filter for random noise and a threshold-based method to detect and correct impulse noise from signal fluctuations or barriers.
  • Multi-Source Information Fusion: A hierarchical particle filter (PF) combined with an extended Kalman filter (EKF) is used. The PF handles trajectory adjustment by generating and weighting particles representing possible pedestrian poses. The EKF estimates the wall parameters (slope and intercept of the wall model). This PF-EKF architecture creates a symbiotic relationship where the PF refines pedestrian poses, and the EKF uses these refined poses to improve wall parameter estimation, which in turn aids the PF's weighting.

Datasets

The experiments were conducted in practical indoor scenarios at a university classroom building. Two specific routes were designed: an L-shape route (192m) and a Z-shape route (136.8m). The device (smartphone with radar) repeated each route 20 times. No specific pre-existing datasets were mentioned; the data was collected by the researchers for this study.

Key Insights and Ideas

  • Radar as a viable indoor PDR aid: The paper demonstrates that an FMCW radar, despite its inherent noise in complex indoor environments, can provide valuable information for improving PDR accuracy.
  • Adaptive operation based on environment: The "operation indicator" is a smart way to handle the non-uniformity of indoor environments. By distinguishing between flat wall segments and corners, the system avoids making potentially detrimental adjustments in challenging areas.
  • Two-stage noise correction is essential: Simply applying a standard Kalman filter isn't enough for radar ranging in dynamic indoor settings. The combination of a Kalman filter for random noise and a threshold-based method for impulse noise is key to obtaining reliable wall distance data.
  • Synergistic PF-EKF for dual estimation: The tight coupling of a particle filter for trajectory and an extended Kalman filter for wall parameters creates a powerful feedback loop. This allows for continuous refinement of both the pedestrian's path and the environmental model, leading to more robust performance.

Key Results

  • Wall Distance Correction: The proposed wall distance correction (DC) module significantly reduced the range of wall distance errors. In the L-shape route, it reduced errors by 30.7% (compared to KF without outlier correction) and by an impressive 93.2% (compared to original observations). For the Z-shape route, these figures were 40.7% and 96.6%, respectively.
  • Wall Parameter Estimation: The RadarPDR scheme, which includes both initialization and updating phases for wall parameters, showed better performance than a scheme with only initialization. The Wall Distance Prediction Error (WDPE) was consistently lower in RadarPDR, especially in later segments of the trajectory, indicating more accurate wall modeling.
  • Overall Trajectory Accuracy: When compared to PurePDR (inertial only), RadarPDR w/o DC, and RadarPDR w/o PU, the complete RadarPDR scheme achieved the best trajectory adjustment performance. It reduced the Average Localization Error (ALE) by 48.1% on the L-shape route and 41.1% on the Z-shape route.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that a single failed trajectory adjustment occurred in one of the 20 tests on the Z-shape route, attributed to a large initial wall parameter error causing EKF divergence.
  • The current setup assumes the radar's ranging direction is perpendicular to the pedestrian's heading, which might not always hold true in all device configurations.
  • Future Work:
    • The authors suggest that the initial direction in the RadarPDR w/o PU scheme can easily deviate from ground truth, implying that further improvements in initialization or parameter estimation could be beneficial.
  • The paper focuses on a specific FMCW radar module. Future work could explore the applicability of the RadarPDR scheme with different types of radar sensors.
  • While the system performs well, further research into more robust handling of extreme indoor environments with very complex layouts or dynamic obstacles could enhance its generalizability.

Summary updated

Abstract

Pedestrian dead reckoning (PDR) is the critical component in indoor pedestrian tracking and navigation services. While most of the recent PDR solutions exploit in-built inertial sensors in smartphones for next step estimation, due to measurement errors and sensing drift, the accuracy of walking direction, step detection, and step length estimation cannot be guaranteed, leading to large accumulative tracking errors. In this paper, we propose a radar-assisted PDR scheme, called RadarPDR, which integrates a frequency-modulation continuous-wave (FMCW) radar to assist the inertial sensors-based PDR. We first establish a segmented wall distance calibration model to deal with the radar ranging noise caused by irregular indoor building layouts and fuse wall distance estimation with acceleration and azimuth signals measured by the inertial sensors of a smartphone. We also propose a hierarchical <i>particle filter</i>(PF) together with an extended Kalman filter for position and trajectory adjustment. Experiments have been conducted in practical indoor scenarios. Results demonstrate that the proposed RadarPDR is efficient and stable and outperforms the widely used inertial sensors-based PDR scheme.
BibTeX
@inproceedings{he2023radarpdr,
  author = {Jianbiao He and Wei Xiang and Qing Zhang and Bang Wang},
  title = {{RadarPDR: Radar-Assisted Indoor Pedestrian Dead Reckoning}},
  booktitle = {Sensors},
  year = {2023},
  doi = {10.3390/s23052782},
}
Figure