Few-Shot Learning for WiFi Fingerprinting Indoor Positioning

Area of Science:

• Computer Science
• Electrical Engineering
• Signal Processing

Background:

• Deep-learning-based WiFi fingerprinting offers accurate indoor location services.
• Existing methods require extensive site surveys and struggle with signal fluctuations.

Purpose of the Study:

• To develop a robust WiFi fingerprinting system using few-shot learning.
• To reduce the need for large labeled datasets and mitigate signal instability.

Main Methods:

• A prototypical network-based positioning system utilizing few-shot learning.
• Temporal Convolutional Network (TCN) for sample embedding and quality assessment.
• Kalman Filter to address spatial ambiguity from signal fluctuations.

Main Results:

• The proposed system establishes a robust RSSI-position matching model with limited labels.
• Achieved superior performance compared to existing deep-learning models using fewer labeled samples.
• Demonstrated effectiveness in mitigating space ambiguity caused by signal fluctuation.

Conclusions:

• The few-shot learning approach significantly enhances WiFi fingerprinting efficiency and robustness.
• The system provides accurate indoor localization with reduced data and improved resilience to signal variability.
• This method offers a promising alternative for practical indoor positioning systems.