[Paper Review] FootSLAM meets Adaptive Thresholding

Johan Wahlström, Andrew Markham+1 more|arXiv (Cornell University)|Nov 1, 2019

Indoor and Outdoor Localization Technologies33 references4 citations

TL;DR

This paper proposes a self-contained, maximum likelihood-based adaptive thresholding method for zero-velocity detection in inertial navigation by leveraging FootSLAM's pose estimates as pseudo-ground truth. By using particle weights from FootSLAM to approximate the likelihood function, the method calibrates the ZUPT threshold without requiring ground truth data, supplementary sensors, or user input, reducing median horizontal error by over 50% compared to fixed-threshold methods under varying gait speeds.

ABSTRACT

Calibration of the zero-velocity detection threshold is an essential prerequisite for zero-velocity-aided inertial navigation. However, the literature is lacking a self-contained calibration method, suitable for large-scale use in unprepared environments without map information or pre-deployed infrastructure. In this paper, the calibration of the zero-velocity detection threshold is formulated as a maximum likelihood problem. The likelihood function is approximated using estimation quantities readily available from the FootSLAM algorithm. Thus, we obtain a method for adaptive thresholding that does not require map information, measurements from supplementary sensors, or user input. Experimental evaluations are conducted using data with different gait speeds, sensor placements, and walking trajectories. The proposed calibration method is shown to outperform fixed-threshold zero-velocity detectors and a benchmark using a speed-based threshold classifier.

Motivation & Objective

• To address the lack of self-contained, large-scale calibration methods for zero-velocity detection thresholds in unprepared indoor environments without map or infrastructure.
• To eliminate dependency on ground truth data, supplementary sensors, or user input for threshold calibration in ZUPT-aided inertial navigation.
• To develop a method that adapts to dynamic factors such as gait speed, sensor placement, and walking surface without prior calibration.
• To enable real-time, autonomous calibration by treating FootSLAM's output as a reliable proxy for ground truth during threshold optimization.

Proposed method

• Formulate the threshold calibration as a maximum likelihood estimation problem using the likelihood of odometry sequences.
• Approximate the likelihood function using particle weights generated by the FootSLAM algorithm during pose estimation.
• Use the FootSLAM output to evaluate the likelihood of different threshold values without requiring external ground truth.
• Perform threshold optimization by selecting the value that maximizes the likelihood of consistent hexagon transitions in the trajectory.
• Ensure convergence of FootSLAM during calibration to maintain stable and consistent walking patterns for reliable estimation.
• Integrate the calibration process into the navigation pipeline to enable online, adaptive thresholding.

Experimental results

Research questions

• RQ1 Can zero-velocity detection thresholds be calibrated without ground truth data, supplementary sensors, or user input?
• RQ2 Can FootSLAM's pose estimates serve as a reliable proxy for ground truth in threshold calibration?
• RQ3 Does the proposed method outperform fixed-threshold and speed-based threshold classifiers under varying gait speeds and sensor placements?
• RQ4 Can the method adapt to changes in gait mode, walking surface, or sensor placement without re-calibration?
• RQ5 Is it possible to achieve stable and accurate calibration using only the internal consistency of FootSLAM's trajectory estimation?

Key findings

• The proposed adaptive thresholding method reduced the median horizontal position error by over 50% compared to a fixed-threshold detector under varying gait speeds.
• The method outperformed a benchmark speed-based threshold classifier across diverse walking trajectories and sensor placements.
• The calibration process achieved stable performance without requiring ground truth data, supplementary sensors, or user input.
• The method demonstrated robustness to variations in gait speed, sensor placement, and walking surface due to its self-calibrating nature.
• FootSLAM's convergence during calibration enabled reliable likelihood approximation, validating the use of its output as a pseudo-ground truth.
• The approach enables calibration in unprepared environments, such as emergency scenes or outdoor areas, where infrastructure is unavailable.