HiMeter: Telling You the Height Rather than the Altitude

Tao Gu, Haibo Ye, Kai Dong, Tao Gu 0001Sensors 2018 · DOI

TL;DR

This paper introduces HiMeter, a novel smartphone application that accurately estimates a user's height (vertical distance to the ground) using only the device's low-power barometer. Unlike traditional altitude measurements, HiMeter focuses on relative changes and avoids GPS or server reliance, offering a practical solution for various mobile applications.

What Problem Does This Paper Address?

Smartphones are ubiquitous, and with them comes a growing demand for context-aware applications. While altitude (distance to sea level) is often provided, it's frequently less useful than "height" (distance to the ground) in many real-world scenarios. For instance, knowing your floor in a building or how high you've climbed a hill is more relevant than your absolute altitude. Existing methods for determining height are often inaccurate, relying on error-prone GPS or barometric pressure readings combined with map data for ground altitude, or requiring external reference points and calibration. This inaccuracy limits the usefulness of height-dependent applications.

Methods

HiMeter tackles this problem by leveraging the smartphone's barometer, which excels at measuring relative pressure changes. The core idea is to measure the change in altitude from a detected "ground" point, rather than trying to determine absolute altitude. The system involves several key steps:

  • Barometer Data Preprocessing: Raw barometer readings are noisy and affected by weather. HiMeter applies a multi-stage filtering process, including outlier removal, low-pass filtering, and discrete wavelet transform, to smooth the data and reduce high-frequency noise.
  • Weather Noise Filtering: A significant challenge is distinguishing pressure changes due to weather from those caused by user movement. HiMeter uses curve fitting on external meteorological data to predict and reconstruct the weather-induced pressure changes, allowing them to be subtracted from the raw readings.
  • Vertical Moving Mode Detection: To accurately identify the ground and interpret altitude changes, HiMeter classifies the user's movement into three modes: "Indoor Mode" (short duration, weather negligible), "Outdoor Mode" (longer duration, weather significant), and "Traffic Mode" (vehicle movement, treated as noise). This classification is achieved by analyzing patterns in the preprocessed barometer data, specifically focusing on the "areas" formed by the readings over time. A hierarchical clustering algorithm (CURE) is used to segment the data, followed by a decision tree classifier to categorize these segments into the three modes.
  • Ground Distinguishing: The system identifies the ground point using two context-inferring rules:
  • When a user transitions from "Traffic Mode" to "Indoor" or "Outdoor Mode," the point at the end of the traffic mode is considered the ground.
  • If a user moves downwards and remains at a stable altitude for at least 10 minutes before moving upwards again, that stable period is identified as the ground.
  • Height Calculation: Once the ground point is identified, the height is calculated by summing the altitude changes from that ground point to the user's current position.

Datasets

  • Online Survey Data: A nation-wide online survey was conducted in China with 1000 valid responses to confirm user desirability for height information and acceptable power consumption.
  • Field Study Data: Data was collected from 10 participants over six days in Nanjing, China. This data was used for training and evaluating the HiMeter system. Participants used their own smartphones, and ground truth was manually measured using a portable infrared range finder and user notes. An electronic barometer sensor was also deployed for weather noise filtering.
  • External Meteorological Data: Data from a weather service provider (Moji Weather) was used for filtering weather-induced pressure changes.

Key Insights and Ideas

  • Height vs. Altitude: The paper strongly corroborates the intuition that "height" (relative to the ground) is often more useful than "altitude" (relative to sea level) for mobile applications.
  • Barometer's Relative Accuracy: The key insight is that while barometers have poor absolute accuracy, their relative accuracy is excellent. This makes them ideal for measuring altitude changes, which is precisely what's needed for height estimation.
  • Moving Mode as Context: Classifying user movement into distinct modes (Indoor, Outdoor, Traffic) is crucial for accurately interpreting barometer readings, filtering noise, and identifying the ground.
  • Ground Identification Rules: The development of simple yet effective rules for identifying the ground point based on traffic mode transitions and periods of stable altitude is a clever way to avoid complex localization or external references.
  • Low-Power, Offline Operation: The ability to perform height calculation solely on the smartphone using a low-power sensor without relying on GPS or server communication is a significant practical advantage.

Key Results

  • Indoor Accuracy: HiMeter achieves an accuracy of within 5 meters in 90% of indoor cases.
  • Outdoor Accuracy: HiMeter achieves an accuracy of within 10 meters in 83% of outdoor cases.
  • Moving Mode Detection Accuracy: The decision tree classifier achieved accuracies of approximately 91% for Indoor Mode, 88% for Outdoor Mode, and 90% for Traffic Mode.
  • Ground Distinguishing Accuracy: The system had a false negative rate of 20% for ground distinguishing.
  • Power Consumption: The estimated extra daily power consumption for HiMeter is approximately 1.65%, which was deemed acceptable by users.
  • Comparison to Existing Works: HiMeter outperformed a common barometer-based approach, a U.S. patent approach, and Liu's barometer-based approach in terms of height accuracy, especially indoors.

Limitations and Future Work

  • Mountainous Areas: HiMeter does not perform as well in mountainous cities, suggesting a limitation in its current ground-distinguishing capabilities in highly varied terrain.
  • Future Work: The authors plan to:
  • Conduct further data collection with more diverse users and locations.
  • Develop a crowdsensing version of HiMeter for Android, incentivizing users to upload data for improved ground-distinguishing patterns.
  • Explore applications for providing altitude change information, which the system handles well even with limitations in ground identification.

Summary updated

BibTeX
@article{gu2018himeter,
  author = {Tao Gu and Haibo Ye and Kai Dong and Tao Gu},
  title = {{HiMeter: Telling You the Height Rather than the Altitude}},
  journal = {Sensors},
  year = {2018},
  doi = {10.3390/s18061712},
}

Loading related content...

Figure