Crowdsourced Indoor Positioning with Scalable WiFi Augmentation.

Area of Science:

• Computer Science
• Electrical Engineering
• Geomatics Engineering

Background:

• Crowdsourcing WiFi signals for indoor positioning databases is efficient but sensitive to crowd density, leading to accuracy degradation in sparse areas.
• Existing methods struggle with data scarcity, impacting the reliability of indoor positioning systems.

Purpose of the Study:

• To propose a scalable WiFi fingerprint augmentation method to improve indoor positioning accuracy.
• To address the limitations of crowdsourced data by generating virtual reference points and modeling spatial WiFi signals.

Main Methods:

• Developed a two-module augmentation method: virtual reference point generation (VRPG) and spatial WiFi signal modeling (SWSM).
• Proposed globally self-adaptive (GS) and locally self-adaptive (LS) approaches within VRPG to identify potential unsurveyed reference points (RPs).
• Employed a multivariate Gaussian process regression (MGPR) model to estimate WiFi signal distributions and predict signals for new RPs.

Main Results:

• Combining GS and MGPR improved positioning accuracy by 5%–20% with halved computation complexity compared to conventional methods.
• Combining LS and MGPR reduced computation complexity by 90% while still offering moderate accuracy improvements.
• Evaluations on an open-source dataset demonstrated the effectiveness of the proposed augmentation techniques.

Conclusions:

• The proposed scalable WiFi fingerprint augmentation method effectively enhances indoor positioning accuracy.
• The VRPG and SWSM modules, particularly with MGPR, offer significant improvements in both accuracy and computational efficiency.
• The adaptive approaches (GS and LS) provide flexibility in balancing accuracy gains and computational load for WiFi-based indoor positioning.