Accurate Localization Method Combining Optimized Hybrid Neural Networks for Geomagnetic Localization with...

Area of Science:

• Geomatics Engineering
• Computer Science
• Artificial Intelligence

Background:

• Location-based services (LBS) offer significant economic and social advantages.
• Geomagnetism presents a ubiquitous, low-cost, and accessible resource for localization.
• Existing geomagnetic localization techniques face challenges with location ambiguity.

Purpose of the Study:

• To develop a robust and accurate fusion localization algorithm.
• To overcome the limitations of traditional geomagnetic localization methods.
• To enhance heading estimation for improved dead reckoning.

Main Methods:

• A five-dimensional hybrid LSTM (5DHLSTM) neural network was developed for geomagnetic localization, with parameters optimized using particle swarm optimization (PSO).
• An eight-dimensional BiLSTM (8DBiLSTM) algorithm was employed for accurate heading estimation in pedestrian dead reckoning (PDR).
• Fusion localization was achieved by integrating geomagnetic localization with an improved PDR (IPDR) utilizing an extended Kalman filter (EKF).

Main Results:

• The proposed PSO-5DHLSTM-IPDR method demonstrated improved localization accuracy in experimental validation.
• The algorithm exhibited good robustness and flexibility across different scenarios.
• Enhanced heading accuracy was achieved through the 8DBiLSTM algorithm, contributing to overall localization performance.

Conclusions:

• The developed fusion localization algorithm effectively addresses location ambiguity in geomagnetic localization.
• The integration of PSO-optimized 5DHLSTM and 8DBiLSTM with EKF-based IPDR offers a superior solution for LBS.
• The method shows significant potential for practical applications requiring precise and reliable positioning.