Forward collision warning (FCW) systems have been widely used in trucks to alert drivers of potential road situations so they can reduce the risk of crashes. Research on FCW use shows, however, that there are differences in drivers’ responses to FCW alerts under different scenarios. Existing FCW algorithms do not take differences in driver response behavior into account, with the consequence that the algorithms’ minimum safe distance assessments that trigger the warnings are not always appropriate for every driver or situation. To reduce false alarms, this study analyzed truck driver behavior in response to FCW warnings, and k-means clustering was adopted to classify driver response behavior into three categories: Response Before Warning (RBW), Response After Warning (RAW), and No Response (NR). Results showed that RBW clusters tend to occur at long following distances (>19 m), and drivers applied braking before the warning. In RAW clusters, deceleration after warning is significantly more forceful than before warning. NR clusters occur at short distances, and deceleration fluctuates only slightly. To optimize the FCW algorithm, the warning distance was divided into reaction distance and braking distance. The linear support vector machine was used to fit the driver reaction distance. The long short-term memory method was used to predict braking distance based on each of the three response scenarios: R was 0.896 for RAW scenarios, 0.927 for RBW scenarios, and 0.980 for NR scenarios. Verification results show that the optimized truck FCW algorithm improved safety by 1 % to 5.1 %; accuracy reached 97.92 %, and the false alarm rate was 1.73 %.

中文翻译：

---

考虑卡车驾驶员反应行为特征的前向碰撞预警算法优化

前方碰撞警告 (FCW) 系统已广泛应用于卡车，以提醒驾驶员潜在的道路状况，从而降低碰撞风险。然而，对 FCW 使用情况的研究表明，在不同场景下，驾驶员对 FCW 警报的反应存在差异。现有的 FCW 算法没有考虑驾驶员响应行为的差异，因此触发警告的算法的最小安全距离评估并不总是适合每个驾驶员或情况。为了减少误报，本研究分析了卡车驾驶员响应FCW警告的行为，并采用k均值聚类将驾驶员的响应行为分为三类：警告前响应（RBW）、警告后响应（RAW）和无响应（NR）。结果显示，RBW 集群往往发生在长跟车距离（>19 m）处，并且驾驶员在发出警告之前就采取了制动措施。在 RAW 集群中，警告后的减速明显比警告前更有力。NR 簇发生在短距离处，减速度波动很小。为了优化FCW算法，将警告距离分为反应距离和制动距离。使用线性支持向量机来拟合驾驶员反应距离。使用长短期记忆方法根据三种响应场景中的每一种来预测制动距离：RAW场景的R为0.896，RBW场景的R为0.927，NR场景的R为0.980。验证结果表明，优化后的卡车FCW算法安全性提升1%~5.1%；准确率达到97.92%，误报率为1.73%。