Feature analysis for drowsiness detection based on facial skin temperature using variational autoencoder : a preliminary study

ABSTRACT

Technology to detect signs of drowsiness in drivers is essential even in the age of automated driving to prevent traffic accidents. In this study, facial skin temperature, which can be measured remotely using infrared thermography, as a measure for determining drowsiness was in focus. Facial skin temperature is an autonomic nervous system index that depends on skin blood flow. It is known that facial skin temperature changes depending on the physiological and psychological state, and that it is affected by drowsiness. We focused on an anomaly detection algorithm called the variational autoencoder (VAE). In this study, a model to detect drowsiness was constructed using VAE with only the facial skin temperature during arousal from sleep and search was made for facial areas where skin temperature fluctuates with drowsiness using the model. As a result, it was found that the side of the nasal dorsum may fluctuate with drowsiness and that facial skin temperature may fluctuate asymmetrically with drowsiness. Skin temperature around the orbit was shown to be an area of possible physiological and psychological significance related to autonomic nervous system activity. Based on the above, the degree of anomaly was confirmed to vary depending on the degree of drowsiness, indicating the usefulness of using VAE for drowsiness detection based on facial skin temperature.

KEYWORDS:

• facial skin temperature
• facial thermal image
• infrared thermography
• drowsiness detection
• deep learning

Disclosure statement

No potential conflict of interest was reported by the author(s).