Indoor positioning with self-adaption parameter optimization based on fingerprints.

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• WiFi fingerprinting is crucial for indoor localization.
• Conventional KNN methods face challenges with high-dimensional noise and parameter selection.
• Existing metaheuristic optimizations like GSA can be computationally complex for large environments.

Purpose of the Study:

• To propose a novel two-stage optimization framework, SA-RLS-GSA-KNN, for enhancing WiFi fingerprinting indoor localization.
• To address the limitations of conventional KNN and computationally intensive metaheuristic approaches.
• To improve positioning accuracy and efficiency in indoor localization systems.

Main Methods:

• Implemented a two-stage optimization: Sparse-Aware Recursive Least Squares (SA-RLS) with L1 regularization for initial AP screening, followed by Gravitational Search Algorithm (GSA) for fine-tuning.
• SA-RLS models RSS-position relationships to reduce the search space by pre-screening critical Access Points (APs).
• GSA optimizes the AP subset and K value using a multi-objective function balancing error, feature count, and computational cost.

Main Results:

• Evaluated on 15 public datasets, including UJIIndoorLoc.
• The proposed SA-RLS-GSA-KNN framework achieved over 20% lower average positioning error compared to the baseline KNN algorithm.
• Demonstrated significant reduction in search space and improved parameter optimization for KNN.

Conclusions:

• SA-RLS-GSA-KNN offers a more accurate and efficient solution for WiFi fingerprinting-based indoor localization.
• The two-stage optimization effectively handles high-dimensional data and complex parameter tuning.
• This framework presents a significant advancement over traditional methods for indoor positioning systems.