Adaptive Multi-Type Fingerprint Indoor Positioning and Localization Method Based on Multi-Task Learning and Weight...

Area of Science:

• Indoor positioning and localization
• Sensor fusion
• Machine learning

Background:

• Received signal strength fingerprinting is unreliable in complex indoor environments.
• Existing methods struggle with accuracy due to environmental variability.
• Integration of multiple sensor data types is needed for robust indoor localization.

Purpose of the Study:

• To propose an adaptive indoor positioning and localization method using multi-type fingerprints.
• To enhance positioning accuracy and reliability in complex indoor environments.
• To leverage multi-task learning (MTL) and Weight Coefficients K-Nearest Neighbor (WCKNN) for improved performance.

Main Methods:

• Integrates magnetic field, Wi-Fi, and Bluetooth fingerprints for positioning.
• Employs multi-task learning (MTL) to fuse features and exploit task synergy.
• Utilizes Weight Coefficients K-Nearest Neighbor (WCKNN) for position prediction and weighted averaging based on error prediction.

Main Results:

• Achieved 98.58% accuracy in classifying locations.
• Demonstrated a mean positioning error of 1.95 meters.
• Validated the effectiveness of the proposed adaptive multi-type fingerprint method.

Conclusions:

• The proposed adaptive multi-type fingerprint method significantly improves indoor positioning accuracy.
• MTL and WCKNN effectively fuse diverse fingerprint data for robust localization.
• This approach offers a reliable solution for indoor positioning challenges.