Methods for Sweat Detection in Cars Based on Thermal Images of the Face

Abstract

As part of automatization of driving, an increasing number of functions in the vehicle interior are being automated. Climate automation is a focus of the developers, since manual operation can lead to distraction from the driving task. This requires information about the vehicle’s surroundings, as well as information about the occupants. An example of such information is the measurement of the occupant’s perspiration. In this work we measure the local temperature variation distribution caused by sweating by thermal imaging and evaluate the resulting heat map by methods commonly applied to characterize surface roughness. We compare different characteristic values of the surface roughness according to EN ISO 25178, as well as the radially averaged power spectral density (RAPSD). Using signal detection theory, we evaluate the robustness of the different measures of roughness of the heat maps of selected areas on the forehead of persons sitting in the driver seat of a test car. The results show that the different methods provide different levels of distinction. The RAPSD value provides the best differentiation with 90% hits and 5% false alarms. If sweating is detected in the vehicle, appropriate countermeasures can then be initiated.