Investigating Driver Behavior in Complex Traffic Situations While Driving Partially Automated Vehicles

摘要

Traffic complexity critically influences driver task demands in partially automated vehicles, yet subjective perception and its behavioral indicators remain underexplored in real-world settings. This paper analyzes driver behavior - vehicle interaction, glance patterns, and guiding fixation - across varying levels of subjective traffic complexity, using real-world data from 20 drivers in real urban traffic. Traffic complexity was determined by expert labeling and served as ground truth for vehicle data. Statistical analysis of 16 driver behavior metrics revealed small but significant trends with increasing complexity: deviation from speed limit increased, brake rate increased while braking intensity decreased, horizontal gaze dispersion and entropy widened, and guiding fixation rate decreased, indicating defensive adaptation and perceptual shifts. Contributions include real-world validation of gaze metrics and guiding fixation under subjective complexity, novel insights from gaze and guiding fixation entropy metrics, and the identification of promising indicators~(driven speed, brake rate, gaze yaw entropy, guiding fixation rate) for complexity-adaptive partially automated vehicles. While based on a limited urban sample and expert-labeled subjective complexity, the findings provide a foundation for combined complexity scores and their integration into complexity-adaptive, partially automated vehicles, boosting human-like automation and enhancing safety and predictability in the traffic system.