Discriminating Between Indoor and Outdoor Environments During Daily Living Activities Using Local Magnetic Field Characteristics and Machine Learning Techniques

Have you ever wondered how your smartwatch or smartphone knows whether you are inside a building or outside in the open air? Many modern applications, from fitness tracking to navigation systems, rely on this distinction to function properly. Traditionally, wearable devices use motion sensors like accelerometers and gyroscopes to estimate movement and infer location. However, these methods often struggle in real-world environments, where conditions may vary unpredictably.

This study introduces a new way to improve indoor-outdoor classification by focusing on a different type of sensor: the magnetometer. Unlike motion sensors, which measure acceleration and rotation, a magnetometer detects Earth’s magnetic field. Since buildings, electrical systems, and structural materials alter magnetic fields, analyzing these distortions provides valuable clues about whether a person is indoors or outdoors.

Studying and knowing about these manufactured distortions are essential since many apps and devices like fitness trackers, health monitors, and navigation tools need to determine whether the person is indoors or outdoors. For example, doctors studying mobility in elderly patients must account for uneven outdoor terrain that might affect walking patterns. At the same time, fitness apps adjust calculations based on whether someone is running indoors on a treadmill or outdoors.

Current methods rely on GPS, Wi-Fi, or Bluetooth, but these have drawbacks:
• GPS drains battery life and doesn’t work well indoors.
• Wi-Fi/Bluetooth raises privacy concerns (e.g., tracking user locations).

Motion-based inertial data, such as accelerometer signals, also have limitations. For instance, indoor and outdoor running may generate similar patterns, making it difficult to differentiate between them. However, the developed method improves classification accuracy and reliability by incorporating magnetometer data, even in complex environments. Magnetometers are already built into many wearables, are energy-efficient,
and do not compromise user privacy.

To develop the system, the researchers collected data from wearable sensors and compared machine and deep learning techniques to analyze patterns in magnetic field variations. The model was trained on real-world data rather than relying on predefined rules to recognize key differences between indoor and outdoor settings.

Unlike traditional classification methods, this method can adapt to different environmental and geographical locations, thus making it more robust and scalable. This system was also tested on diverse patient data, ensuring authenticity and freedom from specific movement patterns.

The results show that magnetometer-based classification significantly outperforms traditional motion-sensor-based methods, achieving 91% accuracy. The approach demonstrated higher accuracy and better adaptability even in challenging environments, such as urban areas with mixed indoor-outdoor transitions. This research paves the way for smarter, more context-aware applications by refining how wearable devices interpret environmental signals.

This research will have significant implications across various industries, particularly in advancing personal wearable devices. This method offers a low-power, privacy-conscious, and accurate alternative to GPS by utilizing the subtle magnetic differences between indoor and outdoor spaces combined with lightweight AI technology. Additionally, the team has publicly shared its data and code, encouraging further advancements in smart wearable technology.