TL;DR
This paper introduces "FloorPair," a novel system that achieves near 100% accurate floor localization on smartphones using only barometric pressure data. By constructing a "relative pressure map" and employing an iterative optimization technique, FloorPair overcomes the limitations of sensor drift and environmental noise, making indoor positioning more reliable and cost-effective.
What Problem Does This Paper Address?
Accurate indoor positioning is crucial for many smartphone applications, but determining a user's exact floor in a multi-story building has been a persistent challenge. Existing methods often struggle with barometer inaccuracies like sensor drift and temporal variations, leading to unreliable floor detection. While some approaches use other sensors like accelerometers or Wi-Fi, none have consistently achieved the near-perfect accuracy required for critical applications like emergency services or worker tracking. This paper aims to bridge this gap by proposing a robust and highly accurate floor localization system that leverages the ubiquitous barometric sensors found in modern smartphones.
Methods
The core of the FloorPair system lies in its novel approach to building a "relative pressure map." Instead of relying on absolute pressure readings, which are prone to errors, FloorPair focuses on the differences in pressure between floors.
- Pressure-Pairing: The system collects pressure readings from smartphones and pairs them up. These pairs represent the pressure difference between two floors.
- Marginalizing Uncertainty: To combat sensor drift and unreliable absolute pressure values, FloorPair uses a clever technique of pairing pressure values from two floors. This effectively cancels out common drifts and errors, leaving a more accurate relative pressure difference.
- Relative Pressure Map Construction: These pressure pairs are aggregated into a one-dimensional array, forming a "relative pressure map" specific to a building. This map essentially encodes the expected pressure difference between any two floors.
- Iterative Optimization (EM Algorithm): To maintain accuracy over time and adapt to changing environmental conditions (like weather), FloorPair employs an iterative optimization method inspired by the Expectation-Maximization (EM) algorithm. This process continuously refines the reference pressure, ensuring the system remains accurate even as atmospheric pressure fluctuates.
- Entrance Detection: The system also incorporates a method (referencing prior work [13]) to detect when a user enters a building, which is crucial for establishing an initial reference pressure.
Datasets
The paper utilizes several datasets for its experiments:
- Pressure readings from various smartphones: Data was collected from LG V40, LG V10, and Samsung Note 5 devices under different weather conditions (high and low pressure).
- Real-world building data: Extensive field experiments were conducted in various buildings across Korea, including skyscrapers, commercial buildings, and residential apartments. The paper lists 11 different buildings used for testing.
- Publicly available weather data: Pressure variation data from Seoul in February 2019 was used to analyze temporal pressure changes.
Key Insights and Ideas
- Relative Pressure is Key: The fundamental insight is that while absolute barometric pressure is noisy and unreliable for floor localization, the difference in pressure between adjacent floors is remarkably consistent and can be reliably measured.
- Pairing for Robustness: By pairing pressure readings from two floors, the system effectively cancels out common sensor drifts and environmental noise, leading to a more accurate representation of the pressure differential.
- Iterative Refinement for Sustainability: The use of an EM-like iterative optimization process is crucial for long-term accuracy. It allows the system to adapt to gradual changes in atmospheric pressure due to weather, preventing drift from accumulating and degrading performance over time.
- Cost-Effectiveness: The method requires minimal data collection (only a few pressure pairs) and relies on readily available smartphone sensors, making it a low-cost solution for widespread deployment.
Key Results
The FloorPair method demonstrated impressive results in extensive field experiments:
- Near 100% Accuracy: The system achieved 100% accuracy in most of its experimental trials across various buildings and smartphone models. This accuracy was maintained even over extended periods (several months).
- Robustness Across Devices: The method's effectiveness was validated across different smartphone models (Samsung Galaxy Note 9, V40 ThinQ, V10, Galaxy Note 5), indicating its generalizability.
- Minimal Data Requirements: The paper highlights that for buildings with uniform floor heights, a single floor pair is sufficient to construct a relative pressure map and achieve 100% accuracy. For buildings with varying floor heights, a few pairs are still sufficient for accurate mapping.
Limitations and Future Work
- Extreme Weather Conditions: The authors acknowledge that the system's 100% accuracy might not hold under extreme weather conditions like tornadoes or hurricanes, where pressure fluctuations are exceptionally large.
- Continuous Vertical Movement: For users who move vertically for extended periods (e.g., continuously using an elevator for 5 minutes), the system might experience slight errors (around 0.017% error).
- Slow Vertical Movement: If a user moves vertically very slowly (below 12 cm/s), the system might misinterpret this as a weather change, potentially leading to erroneous updates of the reference pressure.
- Future Work: The authors plan to augment their method with recent artificial intelligence techniques to further enhance its adaptability to diverse environments. They also suggest that the EM algorithm's framework makes it amenable to applying advanced machine learning techniques for floor localization.
Summary updated