A High-Precision Magnetic-Assisted Heading Angle Calculation Method Based on a 1D Convolutional Neural Network (CNN)...

Area of Science:

• Geophysics
• Computer Science
• Robotics

Background:

• Geomagnetic field data is valuable for indoor pedestrian navigation due to its ubiquity and independence from artificial signals.
• Complex indoor magnetic disturbances significantly degrade the accuracy of magnetic-assisted navigation systems.
• A critical need exists for methods to filter out reliable geomagnetic data for high-accuracy indoor inertial navigation.

Purpose of the Study:

• To propose and evaluate an algorithm for screening undisturbed geomagnetic field data for indoor navigation.
• To improve the accuracy of heading angle calculations in pedestrian inertial navigation systems.

Main Methods:

• An algorithm based on a one-dimensional convolutional neural network (1D CNN) was developed to screen magnetic field data.
• Magnetic data within a time window was encoded into a time series, and a 1D CNN with two convolutional layers extracted features.
• Unsupervised clustering in the feature space was employed to classify magnetic data, avoiding reliance on artificial labels.

Main Results:

• The proposed method effectively distinguished between clean geomagnetic data and disturbed indoor magnetic data.
• Significant improvements in the accuracy of heading angle calculations were achieved.
• The algorithm demonstrated its capability to enhance the performance of indoor pedestrian navigation systems.

Conclusions:

• The 1D CNN-based algorithm provides a robust solution for filtering magnetic field data in challenging indoor environments.
• This approach offers a viable technical pathway towards realizing high-precision indoor pedestrian navigation.
• The unsupervised classification method ensures reliable data screening without the need for manual labeling.