Driver Anxiety Detection Based on Seated Pressure Characteristics and Identification of Anxiety-Inducing Scenarios

Driving anxiety is a major issue that compromises the safety and experience of driving. It has been demonstrated that negative emotions with a high arousal factor like anxiety are manifested in body posture and sitting behavior. This paper will investigate one of the ways of identifying anxiety by assessing the pressure distribution in the sitting posture, and discuss driving situations that have a strong correlation with causing anxiety. Thirty people were recruited through a campus social media platform. The experimental design was a one-factor within-subject experimental design in which the researcher used standardized audio materials and a digital countdown task as a means (or inducement) of achieving calm (baseline) and anxiety, respectively. The induction effects were validated using the Self-Assessment Measure (SAM). Also, a pathway accommodating eight driving conditions was established to address the depth of pressure distribution in each condition by means of pressure mats to examine the behavior of the subjects in the relaxed and anxious conditions. The evaluation of both subjective and objective data was performed using the Wilcoxon signed-rank test, and at the same time, we explored the relationships that existed among the driving situations and anxiety levels. The research findings reveal the following: (1) Compared to baseline emotional state, anxiety-induced conditions exhibit heightened pressure distribution and increased volatility in the thigh, hip, and lower back regions, accompanied by greater anterior-posterior center-of-gravity sway. (2) The study identified 40 significant features distinguishing anxiety from calmness, including aTHR_Max and rCOPBTL_Std, primarily distributed across the left leg, right hip, and lower back regions. (3) Through baseline correction and cosine similarity analysis, scenarios prone to triggering anxiety were identified as those involving high uncertainty and high interactivity (e.g., traffic congestion and entering roundabouts); scenarios characterized by continuity and high predictability (e.g., consecutive turns and parking) showed weaker associations with anxiety. This study provides new data support and design rationale for in-vehicle emotion recognition systems and emotion-intervention-based human-machine interaction design.