Adaptive Integrated Particle Filter for Lightweight Matching Localization Based on Magnetic Map and Inertial Sensors

Magnetic map matching localization has various strategic applications, such as drone reconnaissance, submarine navigation, cruise missile guidance, etc. However, the complexity of matching localization architectures based on magnetic map often fails to meet the fast position update frequency demands for high-speed moving targets. To address this issue, this article proposes a lightweight matching localization architecture, i.e., an adaptive integrated particle filter (AIPF) method based on magnetic map and inertial sensors. The AIPF method includes the prediction phase, correction phase, normalization phase, integration phase, and estimation phase. By pre-establishing a Bayesian probability integral table (BPIT) based on standard normal probability distribution functions and expectation functions, the integration results can be directly queried from the BPIT adaptively, avoiding the time consuming integration operation when matching the potential magnetic fingerprint. The heading and step length can be obtained by dead reckoning based on inertial sensors. Extensive experiments show that when the heading uncertainty is less than 1.2°, or the step length uncertainty is less than 30% of the precision of reference magnetic map, the AIPF method has comparable positioning accuracy while the computational complexity is reduced to one-tenth of that in the compared methods, such as particle filter, the adaptive optimization firefly algorithm, extended Kalman particle filter, etc. Moreover, when the experimental condition is more challenging, the AIPF method demonstrates better robustness than the compared methods. This work provides a new solution for lightweight matching localization algorithms using magnetic maps and inertial sensors, suitable for various applications requiring high position update frequencies or fast target localization and tracking.