Browsing by Author "Doddabasappla Basavarajappa, Kruthi"

Now showing 1 - 1 of 1
  • Thumbnail Image
    ItemOpen Access
    Cough Event Recognition Using Signal-Processing Based Feature Sets and Machine Learning, with Tri-Axial Accelerometer Sensor Worn at Multiple Body Points
    (2021-12-22) Doddabasappla Basavarajappa, Kruthi; Vyas, Rushi J.; Murari, Kartikeya; Yanushkevich, Svetlana N; Medeiros de Souza, Roberto
    Human activity recognition (HAR) from time-series accelerometer and gyroscope sensor data has seen tremendous progress in recent years. Laying, standing, sitting, walking, walking down, and walking upstairs are the daily human activities that are commonly classified using sensor data. Cough is a common human activity and is also a symptom of various diseases including the novel coronavirus disease 2019 (COVID-19). Cough detection and classification are well investigated in recent literature to varying levels of success. But, in most of the studies, sensor data for cough activity is collected during still conditions such as sitting or standing and from a specific location such as chest and neck only. The body position of data recording considerably impacts the data, significantly affecting the classification accuracy. In our study, we place tri-axial accelerometer sensors at different spots on the human body where a smartphone or wearable device such as earphones or headphones are commonly worn. We studied the data with statistical and machine learning (ML) based signal processing methods to find the best accelerometer sensor position to detect coughing events accurately on the human body. Our study finds the most suitable sensor position for cough recognition considering the noise introduced by walking and considering different human heights. The proposed multi-band frequency-domain features such as Spectral Summation, Spectral Maximum, and Spectral Spread of acceleration signal offer higher classification accuracy for cough activity.