Robust Stride Detector from Ankle-Mounted Inertial Sensors for Pedestrian Navigation and Activity Recognition with...

Area of Science:

• Biomedical Engineering
• Computer Science
• Robotics

Background:

• Pedestrian trajectory reconstruction is crucial for applications like activity recognition and navigation.
• Existing methods struggle with atypical gaits and challenging environments.
• Ankle-mounted inertial devices offer a practical solution for capturing gait data.

Purpose of the Study:

• To develop and validate a novel stride detector algorithm for accurate pedestrian trajectory reconstruction.
• To improve the detection of both normal and atypical walking strides.
• To enhance the robustness of trajectory reconstruction in diverse and challenging conditions.

Main Methods:

• A stride detector algorithm combined with zero velocity update (ZUPT) inspired technique.
• Sensor alignment and machine learning for stride detection and trajectory analysis.
• Adaptive window inference instead of fixed-size sliding windows.

Main Results:

• 100% detection of normal walking strides and over 97% detection of atypical strides (small steps, side steps, backward walking).
• High robustness in critical situations (sitting, bicycling) with less than two false strides per hour.
• Over 99% success rate in healthy adults and children, and over 97% in children with movement disorders.

Conclusions:

• The proposed algorithm significantly outperforms existing methods for pedestrian trajectory reconstruction in daily life contexts, especially indoors.
• The computed stride trajectories provide essential information for robust activity recognition across various gaits.
• This adaptive, machine learning-based approach offers a promising solution for accurate and reliable gait analysis.