Detecting anger-provoking events with smartphones: a naturalistic driving study

Implementing real-time driving anger detection and appropriate intervention are crucial for road safety. A naturalistic driving study (NDS) was conducted to further validate the anger detection method tested in simulated experiments. Thirty-four drivers participated in the tests, each lasting one to two weeks. A smartphone with a self-developed application was used to record the encountered anger-provoking events, drivers' facial expressions, and vehicle kinematic data. Drivers' anger-related traits were collected through questionnaires. A total of 570 events were collected. Abnormal lane-changing, being blocked, and slow driving were the most common anger triggers, mainly occurring during morning rush hours on ring roads and inbound/outbound highways/main roads. The driving anger detection model with the collected data achieved 93.6% accuracy and 93.4% F1 score. Anger-sensitive features and their variations during anger were presented. These findings may enhance drivers' emotional experience in intelligent vehicles and facilitate the development of emotion detection and intervention systems.