Adaptive Pedestrian Stride Estimation for Localization: From Multi-Gait Perspective.

Area of Science:

• * Inertial Navigation and Sensor Fusion
• * Human Motion Analysis
• * Biomechanical Engineering

Background:

• * Pedestrian Dead Reckoning (PDR) systems rely on accurate stride length estimation.
• * Existing methods struggle with individual variations and diverse walking speeds.
• * Individual differences significantly impact stride length calculation accuracy in PDR.

Purpose of the Study:

• * To develop an adaptable stride length prediction strategy for PDR systems.
• * To address challenges posed by individual differences and wide walking velocity fields.
• * To enhance the accuracy of stride length estimation in PDR applications.

Main Methods:

• * A multi-gait division algorithm segments strides into distinct phases (push-off, swing, heel-strike, stance) using IMU data.
• * Correlation analysis identifies relationships between gait phase features and stride length.
• * Merged multi-regression models predict stride length based on gait phase features.

Main Results:

• * Gait segmentation achieved high F-scores (0.748-0.819) and IoU scores (0.482-0.69).
• * The proposed stride length estimation achieved a Root Mean Square Error (RMSE) of 151.933.
• * Relative error for total distance across varying speeds was less than 2%.

Conclusions:

• * The gait phase segmentation algorithm accurately recognizes gait phases across a wide range of walking speeds.
• * The fused stride length estimation method offers improved accuracy without parameter tuning.
• * This approach surpasses traditional full-stride estimation methods in PDR systems.