A Wi-Fi/PDR Fusion Localization Method Based on Genetic Algorithm Global Optimization.

Area of Science:

• Robotics and Autonomous Systems
• Geomatics Engineering
• Signal Processing

Background:

• Indoor localization faces challenges with Extended Kalman Filter (EKF) sensitivity to initial states and nonlinear drift in Wi-Fi fingerprinting and Pedestrian Dead Reckoning (PDR) fusion.
• Traditional fusion methods struggle with multipath noise and require precise initial conditions, limiting their practical application.

Purpose of the Study:

• To develop and evaluate a robust Wi-Fi/PDR fusion localization approach using global geometric alignment optimized by a Genetic Algorithm (GA).
• To overcome the limitations of traditional filters by eliminating dependence on initial states and mitigating noise.

Main Methods:

• Modeling the PDR trajectory as a geometric entity for global search of optimal 2D similarity transformation.
• Aligning the PDR trajectory with discrete Wi-Fi observations using a Genetic Algorithm (GA) for optimization.
• Employing a Huber-based cost function to enhance robustness against outliers.

Main Results:

• The proposed GA fusion achieved a mean error of 0.878 m, significantly outperforming Wi-Fi, PDR, and EKF fusion.
• Demonstrated superior trajectory fidelity (DTW = 0.390 m) and reduced maximum deviation (Hausdorff = 1.904 m).
• 90% of GA fusion errors were within 1.5 m, confirming high accuracy and robustness, even with 30% artificial Wi-Fi outliers.

Conclusions:

• The GA-based fusion method offers a practical and effective solution for high-accuracy indoor localization.
• The global optimization framework and Huber cost enhance robustness against initial condition sensitivity and noise.
• This approach shows significant potential for multi-sensor fusion in real-world indoor positioning systems.