New Position Candidate Identification via Clustering toward an Extensible On-Body Smartphone Localization System

Area of Science:

• Human-Computer Interaction
• Ubiquitous Computing
• Machine Learning

Background:

• On-body device position awareness is crucial for high-quality smartphone services.
• Current systems support only pre-defined positions, misclassifying new ones.
• This limitation hinders usability when users adopt novel device placements.

Purpose of the Study:

• To propose a framework for discovering and incorporating previously unsupported on-body device positions.
• To develop a method for identifying new position candidates from user data.
• To reduce user burden in labeling and enable on-the-fly system adaptation.

Main Methods:

• Utilized clustering, specifically Density-Based Spatial Clustering of Applications with Noise (DBSCAN), for grouping similar device positions.
• Developed a novel method to determine the optimal 'eps' parameter for DBSCAN, enhancing cluster accuracy.
• Proposed criteria for qualifying identified clusters as new, previously unrecognized device positions.

Main Results:

• The proposed method effectively identifies new device positions without prior knowledge of their quantity.
• Optimizing the DBSCAN 'eps' parameter significantly improves the accuracy of new position candidate identification.
• The framework successfully prepares reliable datasets for on-the-fly retraining, reducing manual labeling efforts.

Conclusions:

• The developed framework enables adaptive on-body device position awareness systems.
• This approach enhances smartphone service quality by accommodating user-defined device placements.
• The method offers a practical solution for discovering and integrating new positions, improving overall system robustness and user experience.