Automatic Recognition of Emotions and Actions in Bi-modal Video Analysis

Abstract

Driver's psychological state is a major concern in traffic safety and an automatic detection of unusual emotions and actions of motor vehicle driver may help prevent and predict the traffic accidents. In this paper, we first study a number of stimulation methods to induce driver's emotions and actions in a simulated driving experiment. We adopt video games, noise stimulation, sleep deprivation, etc., to induce negative emotions including fidgetiness, anger, tiredness, and anxiety. Potentially dangerous actions, such as looking away, texting with a phone, talking on a phone, etc., is posed by volunteer subjects. Using the recorded video data and speech data we established a basic emotion and action database. Second, we propose an efficient multi-modal emotion and action recognition framework, in which facial expression recognition, speech emotion recognition, and action recognition are combined and fused for improved performance.