Heart Rate Extraction Based on Near-Infrared Camera: Towards Driver State Monitoring

In this paper, a remote sensing method for heart rate (HR) detection is proposed via near-infrared facial video data, which can be used for the online monitoring of the physiological parameters of the drivers and further analysis so as to enhance the driving safety. The proposed HR detection method is based on an automatic facial tracking algorithm, i.e., the Kanade-Lucas-Tomasi, to transform the facial images into time-series signals. Empirical mode decomposition, bandpass filtering and fast Fourier transformation are applied to the time-series signals for the HR extraction. Indoor experiments in different scenarios are conducted to discuss the involved impact factors, i.e., the distance between the near-infrared camera and the participant, the illumination intensity and the video duration, to acquire the optimal settings for HR measurement. It is more robust and convenient than those of the current commercial devices and webcam-based HR measurement approaches, especially in complex illumination environments, such as wearing hat, glasses or makeup, with measurement accuracy 95%. Field experiments are also performed for the prognosis of physical state of the driver.