PROJECT TITLE :

Leveraging Acoustic Signals for Fine-grained Breathing Monitoring in Driving Environments

ABSTRACT:

The drivers' physical and mental health is critical to the preservation of road safety in this day and age, when people spend longer and longer periods of time behind the wheel. Patterns of breathing are essential indicators of the health of drivers while they are operating motor vehicles. Existing studies on breathing monitoring require active user participation, such as wearing special sensors, or relatively quiet environments during sleep; these conditions are difficult to replicate while driving due to the noise. In this study, we propose a fine-grained breathing monitoring system called BreathListener. This system makes use of the audio devices found on smartphones in order to estimate the fine-grained breathing waveform in environments where driving is prevalent. We were able to determine that the energy spectrum density (ESD) of acoustic signals can be utilized in order to capture breathing procedures in driving environments by conducting research on data that was collected from actual driving environments. BreathListener eliminates interference from driving environments in ESD signals by using background subtraction and variational mode decomposition. This allows the software to extract the breathing pattern from the ESD signals (VMD). After that, the extracted breathing pattern is converted into the Hilbert spectrum. Following that, we design a deep learning architecture based on generative adversarial networks (GAN) to generate fine-grained breathing waveforms from the Hilbert spectrum of extracted breathing patterns in ESD signals. Experiments conducted with ten different drivers in real driving environments demonstrated that BreathListener is capable of accurately capturing the breathing patterns of drivers while they are in those environments.