Autonomous Fingerprinting and Large Experimental Data Set for Visible Light Positioning.

Area of Science:

• Indoor positioning systems
• Optical wireless communication
• Robotics and automation

Background:

• Visible Light Positioning (VLP) offers a promising alternative for indoor localization.
• Traditional data collection for VLP is labor-intensive and limits dataset scale.
• Accurate ground truth is crucial for evaluating VLP algorithms.

Purpose of the Study:

• To develop an autonomous method for collecting Visible Light Positioning (VLP) data.
• To conduct a comprehensive investigation of VLP accuracy using a large experimental dataset.
• To compare the performance of Machine Learning (ML) and channel model-based positioning techniques.

Main Methods:

• An autonomous data collection system utilizing consumer-grade Virtual Reality (VR) tracking for precise ground truth.
• Deployment of a mobile robotic platform equipped with a VLP receiver and VR tracker.
• Collection of Received Signal Strength (RSS) data at 7344 locations within a controlled test environment.

Main Results:

• The VR ground truth system achieved median and 90th percentile errors of 4.24 mm and 7.35 mm, respectively.
• Ridge regression demonstrated the highest accuracy among tested ML techniques.
• ML-based positioning outperformed model-based methods with large datasets, while model-based methods were superior with small datasets.

Conclusions:

• Autonomous data collection significantly enhances the scale and accuracy of VLP experiments.
• Machine learning techniques, especially ridge regression, provide high localization accuracy for VLP with sufficient data.
• The choice between ML and model-based positioning depends on the available data volume for calibration and training.