Multimodal estimation of a driver´s spontaneous irritation

In this paper we present our latest achievements in the continuous estimation of a driver´s spontaneous irritation. Experiments are conducted with data from 20 drivers, recorded under real driving conditions. While driving, participants also interact with a speech dialogue system to retrieve and play music. A fusion method is proposed to integrate information on the driving environment, driver behavior, driver´s physiological state, and speech recognition results. Overall, we are able to correctly detect 80% (true positive rate) of the irritation, and, when drivers are not irritated, we only make mistakes 9% of the time (false positive rate). Results also support the relevance of gas- and brake-pedal operation as well as speech recognition results in irritation estimation.