A high degree of automated driving distracts drivers more easily, resulting in slow recognition of critical events during driving and slow responses to emergencies. Automated driving and manual switching processes are also prone to erroneous decisions. We conducted a simulated automated driving experiment to study participants' visual perception characteristics during the take‐over of vehicle control. The present study used dynamic videos to imitate the driving situations when drivers returned their gaze from the distractive source to the road. We collected the drivers' eye movement data to analyze the search strategy and physiological characteristics of the drivers after the take‐over reminder. The results showed that the instant information search method of the drivers was scanning of the driving scene. When the degree of distraction deepened and the hazard level of scenes increased, the pupil diameter of the drivers increased and the fixation duration became longer. These findings can help to design take‐over warnings and support more intelligent automated driving systems to judge whether measures should be taken to interfere with the driver's operation to avoid collisions. Furthermore, the drivers' fixation point distribution focused on the left side and the lower side of the scene. We suggest that the take‐over warning is displayed in the head‐up display. This study provides a better understanding of drivers' visual perception characteristics when drivers' eyesight returns from other distractors to the driving scene and a good theoretical basis for the design of hazard warning information for automated driving.

中文翻译：

---

研究自动驾驶中车辆控制接管过程中驾驶员的视觉感知特性

高度自动化的驾驶更容易分散驾驶员的注意力，从而导致驾驶过程中对关键事件的识别变慢以及对紧急情况的响应变慢。自动驾驶和手动切换过程也容易做出错误的决定。我们进行了模拟自动驾驶实验，研究了在接管车辆控制过程中参与者的视觉感知特征。本研究使用动态视频来模拟驾驶员将注意力从分散注意力的源头返回道路时的驾驶情况。我们收集了驾驶员的眼动数据，以分析接管提醒后驾驶员的搜索策略和生理特征。结果表明，驾驶员的即时信息搜索方法是对驾驶场景进行扫描。当分散注意力的程度加深并且场景的危险程度增加时，驾驶员的瞳孔直径会增加，注视时间会变长。这些发现可以帮助设计接管警告，并支持更智能的自动驾驶系统，以判断是否应采取措施干扰驾驶员的操作以避免碰撞。此外，驾驶员的注视点分布集中在场景的左侧和下方。我们建议在平视显示器中显示接管警告。这项研究可以更好地理解驾驶员的视力从其他干扰物返回到驾驶场景时的视觉感知特性，并为设计用于自动驾驶的危险警告信息提供了良好的理论基础。