LightGBM Indoor Positioning Method Based on Merged Wi-Fi and Image Fingerprints

Area of Science:

• Computer Science
• Electrical Engineering
• Geomatics Engineering

Background:

• Indoor positioning is a significant challenge.
• Existing Wi-Fi fingerprinting methods offer limited accuracy (2-10m).
• High-precision indoor positioning is crucial for many applications.

Purpose of the Study:

• To develop an advanced indoor positioning algorithm.
• To enhance positioning accuracy by fusing Wi-Fi and visual data.
• To improve upon existing single-source fingerprinting techniques.

Main Methods:

• A hybrid fingerprinting approach combining Wi-Fi and visual information.
• Feature extraction using Kernel Principal Component Analysis (KPCA) for Wi-Fi data.
• Feature extraction using Local Binary Patterns (LBP) for visual data.
• Fusion of Wi-Fi and visual features for merged-fingerprint generation.
• Utilizing Light Gradient Boosting Machine (LightGBM) for the positioning stage with feature bundling and histogram optimization.

Main Results:

• The proposed algorithm achieved 90% positioning accuracy within 1.53 meters.
• Accuracy improved by over 20% compared to single-fingerprint methods.
• Performance enhancement exceeded 15% compared to other existing methods.
• The average locating error was recorded at 0.78 meters.

Conclusions:

• Combining Wi-Fi fingerprints and visual information offers a superior approach to indoor positioning.
• The LightGBM model with optimized features provides highly accurate and efficient indoor localization.
• This hybrid method addresses the limitations of traditional Wi-Fi fingerprinting for high-precision needs.