Development of a Method to Predict Crash Risk using Trend Analysis of Driver Behavior Changes over Time

<i><b>Objective</b></i>: This study aimed at identifying and predicting in advance the point in time with high risk of virtual accident before a virtual accident actually occurs using the change of behavioral measures and subjective rating on drowsiness over time and the trend analysis of each behavioral measure. <i><b>Methods</b></i>:The behavioral measures such as neck bending angle and tracking error in steering maneuvering during the simulated driving task were recorded under the low arousal condition of all participants who stayed up all night without sleeping. The trend analysis of each evaluation measure was conducted by using a single regression model where time and each measure of drowsiness corresponded to an independent variable and a dependent variable, respectively. Applying the trend analysis technique to the experimental data, we proposed a method to predict in advance the point in time with high risk of virtual accident (in real-world driving environment, this corresponds to a crash) before the point in time when the participant would have encountered a crucial accident if he or she continued driving a vehicle (we call this the point in time of virtual accident). <i><b>Results</b></i>: On the basis of applying the proposed trend analysis method to behavioral measures, we found that the proposed approach could predict in advance the point in time with high risk of virtual accident before the point in time of virtual accident. <i><b>Conclusion</b></i>: The proposed method is one of the promising techniques for predicting in advance the time zone with potentially high risk (probability) of being involved in an accident due to drowsy driving, and for warning drivers of such a drowsy and risky state.

**研究目的**：本研究旨在依托随时间变化的行为指标与嗜睡主观评分，以及各行为指标的趋势分析，在虚拟事故实际发生前，提前识别并预测高虚拟事故风险的时间节点。
**研究方法**：针对所有彻夜未眠的受试者，在低唤醒状态下开展模拟驾驶任务，记录其转向操作过程中的颈部弯曲角度、跟踪误差等行为指标。以时间作为自变量、各嗜睡测量指标作为因变量，构建单回归模型对各项评估指标进行趋势分析。基于对实验数据应用趋势分析技术，本研究提出一种预测方法：可在参与者若继续驾驶将遭遇致命事故的时间节点（本研究将其定义为虚拟事故时间节点）之前，提前预判高虚拟事故风险节点——在真实驾驶环境中，该风险节点对应交通事故。
**研究结果**：将所提出的趋势分析方法应用于行为指标后，本研究发现所提方法可在虚拟事故时间节点之前，提前预测高虚拟事故风险节点。
**研究结论**：所提方法是一项极具应用前景的技术，可提前预判疲劳驾驶引发交通事故的潜在高风险时段，并向驾驶员发出该疲劳高危状态的预警。