Accurate region of interest selection for video based heart rate monitoring for a person driving a car

In this paper, a video based heart rate monitoring system for a person driving a car has been proposed using the face of the driver as region of interest. Challenges such as illumination and motion artifacts are present when the car is moving in real life. We have proposed methods for tackling insufficient illumination and face tracking. The video for heart rate measurement is acquired at a rate of 30 fps with a resolution of 640×480 pixels for different illumination and car speeds. The video is recorded using dash cam. We apply insufficient and irregular illumination compensation algorithm in the preprocessing stage. The forehead is tracked by using black pixel tracking algorithm. The forehead is divided into many smaller blocks for correct region of interest selection based on heart rate measurements. The photoplethysmography (PPG) signal is extracted by separating the video frames into three RGB traces, and considering G channel only for further processing. Fast Fourier Transform is applied to the G channel traces, followed by band pass filtering. The peak frequency of the resultant signal determines the heart rate (converted to beats per minute (bpm)). The average error in the heart rate measurement for our proposed system is 2.734 bpm.