LSTM-Based Zero-Velocity Detection for Robust Inertial Navigation (1807.05275v2)

Abstract: We present a method to improve the accuracy of a zero-velocity-aided inertial navigation system (INS) by replacing the standard zero-velocity detector with a long short-term memory (LSTM) neural network. While existing threshold-based zero-velocity detectors are not robust to varying motion types, our learned model accurately detects stationary periods of the inertial measurement unit (IMU) despite changes in the motion of the user. Upon detection, zero-velocity pseudo-measurements are fused with a dead reckoning motion model in an extended Kalman filter (EKF). We demonstrate that our LSTM-based zero-velocity detector, used within a zero-velocity-aided INS, improves zero-velocity detection during human localization tasks. Consequently, localization accuracy is also improved. Our system is evaluated on more than 7.5 km of indoor pedestrian locomotion data, acquired from five different subjects. We show that 3D positioning error is reduced by over 34% compared to existing fixed-threshold zero-velocity detectors for walking, running, and stair climbing motions. Additionally, we demonstrate how our learned zero-velocity detector operates effectively during crawling and ladder climbing. Our system is calibration-free (no careful threshold-tuning is required) and operates consistently with differing users, IMU placements, and shoe types, while being compatible with any generic zero-velocity-aided INS.

• The paper demonstrates that replacing fixed-threshold zero-velocity detectors with an LSTM model significantly improves detection accuracy by reducing position error by over 34%.
• The methodology leverages a hybrid dataset of laboratory and augmented inertial data to recognize diverse motion patterns, enhancing INS performance in GNSS-denied environments.
• The findings imply that data-driven approaches can offer robust, calibration-free solutions for indoor navigation and emergency response localization.

LSTM-Based Zero-Velocity Detection for Robust Inertial Navigation

The paper presents an advanced method for zero-velocity detection in inertial navigation systems (INS), leveraging Long Short-Term Memory (LSTM) neural networks to enhance localization accuracy. Traditional zero-velocity detectors, which typically rely on fixed thresholds, suffer from reduced reliability when encountering varying motion types. The method introduced in this research aims to address these challenges by utilizing a data-driven approach, improving upon standard threshold-based detectors through machine learning techniques.

Summary of Methodology and Results

Inertial navigation systems use data from body-mounted inertial measurement units (IMUs) to estimate an individual's position relative to a starting point. These systems, crucial in environments where Global Navigation Satellite System (GNSS) signals are unavailable, such as indoor settings, are often hampered by drift. Drift is mitigated through techniques like zero-velocity updates, which use periods of detected zero velocity to correct error accumulation. Conventionally, zero-velocity periods are identified using fixed thresholds but are limited due to the variability in human movements.

The researchers propose replacing these fixed-threshold detectors with an LSTM-based model. The model is trained to recognize stationary periods from raw IMU data, focusing on improving zero-velocity event detection accuracy. Training incorporated a combination of inertia data collected in a laboratory environment and augmented using transformations that simulate real-world changes (e.g., varying IMU orientations and velocities). The model demonstrated a substantial improvement in detection accuracy across multiple motion types, not only reducing the position error by over 34% in various scenarios compared to traditional methods but also proving effective in non-standard motion types, such as crawling or ladder climbing.

Practical and Theoretical Implications

Practically, the paper enhances robustness in applications requiring precise indoor navigation, such as emergency response, where reliable localization data can significantly impact operational safety and efficacy. The flexibility of the model across various users, shoes, and IMU placements suggests potential for broad deployment without requiring recalibration or substantial preprocessing.

Theoretically, the paper contributes to the growing body of literature highlighting the effectiveness of machine learning methodologies in improving classical navigation paradigms. The ability of the LSTM model to reliably detect zero-velocity across diverse motion patterns speaks to the potential for similar models to enhance other signal processing applications within the domain.

Future Directions

As with many AI-based systems, future research could explore the integration of additional sensors within the LSTM-based INS architecture to enhance system accuracy further. The potential addition of multi-sensor fusion and context-aware localization techniques could also provide insights into overcoming current limitations in inertial localization. Expanding evaluation datasets to include a wider range of real-world conditions and environments will validate and refine the model's applicability to practical scenarios.

In summary, the paper proposes a rigorous and effective modification to inertial navigation systems, employing LSTM networks to significantly improve zero-velocity detection, thereby enhancing overall localization fidelity. The innovation sets a foundation for the broader application of machine learning techniques in dynamic signal processing challenges inherent to mobile sensing and navigation systems.