Step Length Estimation for Blind Walkers.

Area of Science:

• Robotics
• Human-Computer Interaction
• Machine Learning

Background:

• Smartphone inertial data aids pedestrian dead-reckoning (PDR) for blind pedestrian mobility.
• Accurate step length estimation is crucial for PDR localization.
• Previous step length prediction models were trained exclusively on sighted individuals.

Purpose of the Study:

• To evaluate the performance of existing step length estimation models on blind pedestrians.
• To develop and validate a retrained model for improved step length prediction in blind individuals.

Main Methods:

• Collected inertial data from blind pedestrians using smartphones.
• Trained and tested machine learning models for step length estimation.
• Compared model performance using data from sighted versus blind walkers.

Main Results:

• Step length estimation models trained on sighted walkers showed poor accuracy when applied to blind walkers.
• Retraining the models with data from blind walkers significantly enhanced prediction accuracy.
• The study highlights gait differences impacting PDR performance.

Conclusions:

• Existing step length estimation models require adaptation for blind pedestrians.
• Retraining machine learning models with specific user group data is essential for accurate PDR.
• Improved PDR accuracy can enhance mobility independence for blind individuals.