Mobile User Indoor-Outdoor Detection Through Physical Daily Activities

Area of Science:

• Computer Science
• Ubiquitous Computing
• Machine Learning

Background:

• Accurate user positioning is essential for smart device services like location-based services (LBS) and seamless indoor/outdoor navigation and localization (SNAL).
• Existing methods often rely on external infrastructure, limiting applicability.

Purpose of the Study:

• To develop an automatic, fast, and accurate method for detecting indoor/outdoor environments based on user daily activities.
• To enable seamless switching between Global Positioning System (GPS) and indoor positioning systems.

Main Methods:

• Utilized six daily user activities (walk, skip, jog, stay, climbing stairs up/down) for environment classification.
• Applied ensemble learning methods, including Random Forest and AdaBoost, on selected activity features.
• Evaluated the approach on the HASC-2016 public dataset.

Main Results:

• Achieved 99% accuracy in detecting environment types (indoor/outdoor).
• Demonstrated a high detection rate and good adaptation for environment recognition.
• The method requires only daily activity data, negating the need for external facilities like Wi-Fi or cell towers.

Conclusions:

• The proposed activity-based method offers a highly accurate and efficient solution for indoor/outdoor environment detection.
• This approach enhances the applicability of user positioning for various upper-layer smart device applications.
• Eliminates reliance on external signals, improving robustness and accessibility.