Smartphone-Based Traveled Distance Estimation Using Individual Walking Patterns for Indoor Localization.

Area of Science:

• Computer Science
• Robotics
• Signal Processing

Background:

• Traditional pedestrian dead reckoning (PDR) indoor localization relies on generalized formulas for step count and stride length.
• Existing methods often use manually designed features from sensor signals, limiting adaptability and efficiency.
• There is a need for more accurate and automated methods for indoor localization using readily available smartphone sensors.

Purpose of the Study:

• To develop a novel indoor localization method using a user's smartphone.
• To learn personalized walking patterns for improved distance estimation.
• To reduce the cost and increase the efficiency of dataset construction for localization models.

Main Methods:

• A hybrid multiscale convolutional and recurrent neural network (CNN-RNN) model was proposed to learn pedestrian velocity from segmented sensor signal frames.
• Distance traveled is estimated by calculating velocity and elapsed time, moving beyond traditional step-based methods.
• A real-time, automatic dataset construction approach was developed using synchronized inertial sensor and GPS data collected outdoors.

Main Results:

• The proposed deep learning model achieved a low distance error of <2.4% and >1.5% in indoor experiments.
• The automatic dataset construction method significantly reduced costs and improved efficiency.
• The model demonstrated applicability to various time-series sensor signal processing tasks.

Conclusions:

• The novel hybrid CNN-RNN approach enables accurate indoor localization by learning personalized walking dynamics.
• The method offers a cost-effective and efficient alternative for dataset creation compared to manual methods.
• This deep learning-based strategy enhances the performance of smartphone-based indoor positioning systems.