Machine Learning Models for Indoor Positioning Using Bluetooth RSSI and Video Data: A Case Study

Area of Science:

• Computer Science
• Robotics
• Electrical Engineering

Background:

• Accurate indoor positioning is crucial for many applications.
• Challenges include signal interference and environmental variability.
• Existing systems often struggle with precision in complex settings.

Purpose of the Study:

• To develop and evaluate a multimodal Indoor Positioning System (IPS).
• To integrate Bluetooth Received Signal Strength Indicator (RSSI) and video data.
• To enhance IPS accuracy using machine learning and ensemble techniques.

Main Methods:

• Implemented a multimodal IPS using Bluetooth RSSI and video imagery.
• Trained independent machine learning models for each data source.
• Applied ensemble learning to combine model predictions.
• Deployed and tested the system in a museum environment with deployment constraints.

Main Results:

• Ensemble models significantly outperformed individual RSSI-based and video-based models.
• Multimodal data integration improved positioning accuracy.
• The system demonstrated effectiveness despite multipath interference, low lighting, and limited beacon infrastructure.

Conclusions:

• Multimodal data fusion with ensemble learning enhances IPS accuracy in complex indoor environments.
• The proposed system offers a robust solution for real-world deployments with practical constraints.
• This approach shows significant potential for future indoor localization applications.