Toward Accurate Position Estimation Using Learning to Prediction Algorithm in Indoor Navigation.

Area of Science:

• Robotics and Automation
• Sensor Fusion
• Indoor Localization

Background:

• Smart navigation and location tracking are crucial for mission-critical indoor scenarios, logistics, medicine, and security.
• Indoor localization is a growing field driven by location-based services.
• Existing inertial measurement unit-based methods suffer from accuracy and consistency issues.

Purpose of the Study:

• To propose a novel position estimation system for indoor localization.
• To address the accuracy and consistency challenges in current indoor localization techniques.
• To develop a system that enhances prediction algorithms using sensor fusion and a learning module.

Main Methods:

• A two-module system comprising a learning module and a position estimation module using sensor fusion.
• An Artificial Neural Network-based learning module integrated with a Kalman filter as the prediction algorithm.
• Utilizing data from a next-generation inertial measurement unit with 3-axis accelerometer and gyroscope.

Main Results:

• The proposed system effectively predicts accurate gyroscope and accelerometer readings from noisy sensor data.
• The learning module continuously monitors and enhances the prediction algorithm's efficiency.
• Experiments demonstrated that the Kalman filter with the learning module achieved a lower root mean square error compared to the traditional Kalman filter.

Conclusions:

• The developed learning-to-prediction model significantly improves indoor localization accuracy.
• The integration of a learning module with a Kalman filter offers a robust solution for noisy sensor data.
• The proposed system shows superior performance in terms of accuracy and consistency for indoor localization applications.