Automatic real-time FACS-coder to anonymise drivers in eye tracker videos

Abstract

Driver's face is a rich source of information for understanding driver behaviour. From the driver's face, one could get an idea of the driver's emotional state and where s/he looks at. In recent years, naturalistic driving studies and field operational tests have been conducted to collect driver behavioural data, which often includes video of the driver, from many drivers driving for an extended period of time. Due to the Data Privacy Act, it is desirable to make the driver video anonymous, while preserving the original facial expressions. This paper describes our attempt to make a system that could do so. The system is a combination of an automatic Facial Action Coding System (FACS) coder based on Active Appearance Models (AAMs), a classifier that analyses local deformations in the AAM shape mesh and a 3D visualisation. The image acquisition hardware is based on a SmartEye eye tracker installed in a vehicle. The eye tracker we used provides a constant image quality independent of external illumination, which is a precondition for deploying the system in a vehicle environment. While the system uses Action Unit (AU) activations internally, the evaluation was done using the six basic emotions.