TL;DR
This paper introduces a novel approach for rapidly updating indoor positioning fingerprint databases using a Unmanned Ground Vehicle (UGV) equipped with Simultaneous Localization and Mapping (SLAM). By integrating SLAM with various Signals of Opportunity (SOP) sensors, the system can automatically build and maintain accurate indoor maps and corresponding fingerprint databases, significantly reducing manual effort and improving positioning accuracy.
What Problem Does This Paper Address?
Indoor positioning systems often rely on fingerprinting, which involves creating a database of signal strengths (like WiFi RSSI) at known locations. However, maintaining this database is a major challenge. Signals are sensitive to environmental changes (e.g., furniture rearrangement, new WiFi access points), requiring frequent recalibration. Traditional manual database creation is labor-intensive, time-consuming, and difficult to update in real-time. This leads to outdated databases and reduced positioning accuracy, hindering the practical application of indoor positioning.
Methods
The core of the proposed solution is the NAVIS platform, a custom-built UGV that combines SLAM with multiple Signals of Opportunity (SOP) sensors.
- SLAM for Mapping and Localization: NAVIS uses a 2D LiDAR-based SLAM system (specifically, an Improved Maximum Likelihood Estimated (IMLE) algorithm) to simultaneously build an indoor map and determine its own precise position within that map. This allows for accurate, real-time localization without prior map information.
- Multi-SOP Data Collection: The UGV is equipped with several sensors to collect various SOPs:
- WiFi: Captures Received Signal Strength Indicator (RSSI) from access points.
- Magnetic Field Sensor: Measures magnetic field intensity.
- Light Sensor: Records ambient light intensity.
- Digital Compass: Provides heading information.
- Fingerprint Database Generation: As NAVIS traverses the environment, it logs the collected SOP data along with its SLAM-derived position. This creates a list of position-SOP pairs.
- Spatial Distance-Mean Filter: For WiFi signals, a filter is applied to average signal strengths within a defined radius around each reference point, making the fingerprint more robust to minor signal fluctuations.
- Coordinate Reference Rectification: The local coordinate system generated by SLAM is transformed into a global coordinate reference using a Four Parameters Similarity Coordinate Transformation (FPSCT) method, ensuring consistency for positioning applications.
- Map and Database Updating: The system can detect changes in the environment (e.g., new obstacles, changes in lighting) and update both the spatial map and the SOP fingerprint database accordingly.
Datasets
The paper utilizes two main sources for evaluation:
- Finnish Geospatial Research Institute (FGI) Building: The primary experimental environment for collecting data and testing the NAVIS platform. This includes corridor sections with and without simulated environmental changes.
- Intel Research Lab-Seattle Dataset: A public dataset used to evaluate the mapping accuracy of the SLAM algorithm in a feature-rich office environment.
No new benchmark datasets were introduced, but the experiments demonstrate the system's capability in a real-world setting.
Key Insights and Ideas
- SLAM as a Game-Changer for Database Maintenance: The integration of SLAM with SOP sensors automates the tedious process of fingerprint database creation and updating. This is a significant step towards practical, dynamic indoor positioning.
- Multi-SOP Fusion Enhances Accuracy: Combining data from WiFi, magnetic field, and light intensity sensors demonstrably improves positioning accuracy compared to using WiFi alone. This highlights the benefit of leveraging diverse environmental signals.
- Environmental Change Detection is Crucial: The system's ability to detect spatial and SOP variations in the environment is key to maintaining an up-to-date and accurate fingerprint database. This is essential for reliable indoor positioning.
- UGV Platform Offers Flexibility: The NAVIS platform's mobility and sensor integration allow for rapid data collection over larger areas with denser sampling points than traditional methods.
Key Results
- Faster Database Creation and Update: The proposed method allows for quicker creation and updating of fingerprint databases with higher sampling frequencies and denser reference points compared to traditional manual methods.
- Improved Positioning Accuracy:
- Using WiFi only, approximately 80% of positioning results were within 3 meters.
- When utilizing miscellaneous SOPs (WiFi, light, magnetic), an un-optimized algorithm improved the mean positioning error by 19% and increased the percentage of results within 3 meters to 90.5%.
- Mapping Accuracy: The NAVIS SLAM mapping results were found to be accurate enough for SOP fingerprint database creation and positioning, with minor deviations (e.g., 0.4m in a specific corridor section) attributed to sensor limitations and environmental features.
- Environmental Change Detection: The system successfully detected simulated environmental changes (e.g., placing objects, turning lights on/off) in the corridor, demonstrating its capability for map updating.
Limitations and Future Work
- Limitations:
- The current system uses an "un-optimized" Weight Quick Selection (WQS) algorithm for positioning, suggesting that further optimization could yield even better results.
- The accuracy of the SLAM mapping can be affected by feature-poor environments and sensor footprint size, which can lead to accumulated positional errors.
- The study acknowledges that the current configuration is a preliminary test, and more complex scenarios are needed.
- Future Work:
- Conducting experiments in more dynamic environments with moving pedestrians to simulate crowded scenarios (e.g., supermarkets).
- Investigating and developing optimized positioning algorithms that leverage miscellaneous SOP sources.
- Further improving the SLAM mapping accuracy, particularly in challenging indoor environments.
Summary updated