Step Length Is a More Reliable Measurement Than Walking Speed for Pedestrian Dead-Reckoning

Area of Science:

• Computer Science
• Robotics
• Signal Processing

Background:

• Pedestrian dead reckoning (PDR) is crucial for indoor and outdoor navigation.
• PDR algorithms estimate user path using inertial sensor data from devices like smartphones.
• Current methods often estimate step length or walking speed, with varying accuracy.

Purpose of the Study:

• To compare the reliability of predicting step length versus walking speed for PDR.
• To analyze the performance of a specific neural network architecture (LSTM + fully connected layers) for these prediction tasks.

Main Methods:

• Utilized a recurrent neural network architecture: Long Short-Term Memory (LSTM) followed by four fully connected layers.
• Trained and evaluated the model on a dataset of inertial data collected from twelve participants.
• Assessed the quality of reconstruction for both step length and walking speed predictions.

Main Results:

• The study found that predicting step length yielded more reliable results compared to predicting average walking speed over each step.
• The chosen LSTM-based architecture demonstrated a higher fidelity in reconstructing step length.

Conclusions:

• Directly predicting step length is a more robust approach for PDR than predicting walking speed.
• This improved reliability in step length estimation can enhance the accuracy of PDR-based navigation systems.