Efficient k-NN Implementation for Real-Time Detection of Cough Events in Smartphones

Area of Science:

• Digital Health
• Biomedical Engineering
• Signal Processing

Background:

• Telemedicine for respiratory care lacks objective cough measurement tools.
• Existing cough detectors are inconvenient and costly.
• Smartphones offer a viable alternative but face challenges in noise and computational demands.

Purpose of the Study:

• To develop a robust and efficient smartphone-based cough detection system.
• To address the computational cost challenge for real-time deployment.
• To enable effective telemedicine for respiratory health.

Main Methods:

• Utilized local Hu moments for robust audio feature extraction in noisy environments.
• Implemented an improved vantage point (vp)-tree structure for accelerated k-nearest neighbors (k-NN) search.
• Optimized vp-tree construction and distance functions for faster computations.

Main Results:

• Achieved classification accuracy exceeding 93% for cough detection.
• Demonstrated an 18x speed-up compared to classic vp-trees.
• Achieved a 560x speed-up over standard k-NN implementations, enabling real-time performance on low-end smartphones.

Conclusions:

• The developed smartphone cough detector is efficient and accurate.
• The system overcomes computational limitations for real-time use.
• This technology facilitates enhanced telemedicine for respiratory conditions.