Unified Probabilistic and Similarity-Based Position Estimation from Radio Observations.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Accurate indoor localization remains a challenge, with existing methods often relying on explicit physical models of radio propagation.
• Traditional position estimation techniques typically provide point estimates, limiting their integration into dynamic systems.

Purpose of the Study:

• To develop a purely similarity-based modeling approach for position estimation, independent of explicit physical assumptions.
• To introduce probabilistic position estimates as continuous probability density functions for enhanced integration into state estimation systems.

Main Methods:

• A kernel-based method compares incoming radio propagation data with reference recordings from known locations.
• Weights are assigned to reference positions based on similarity, followed by density estimation to generate a continuous probability distribution.
• The approach is implemented within a Particle Filter (PF) system for smartphone-based indoor localization.

Main Results:

• The similarity-based model achieved higher accuracy in tracking pedestrians under realistic conditions compared to distance-based models.
• The model demonstrated flexibility, performing effectively with both radio signal strength (RSS) and round-trip time (RTT) measurements.
• Probabilistic position estimates were successfully generated as continuous probability density functions.

Conclusions:

• The proposed similarity-based approach offers a robust and flexible alternative for indoor localization, outperforming traditional methods.
• Its ability to provide probabilistic estimates enhances its utility in recursive state estimation and real-world tracking applications.
• The model's independence from specific radio propagation physics allows for versatile application across different measurement types.