为及时识别、预测车辆的换道行为,综合考虑目标车辆及周边车辆的时空交互关系,结合时间卷积网络(Temporal Convolutional Network,TCN)的时序处理能力和长短期记忆(Long Short Term Memory,LSTM)神经网络的门控记忆机制,构建了基于TCNL...为及时识别、预测车辆的换道行为,综合考虑目标车辆及周边车辆的时空交互关系,结合时间卷积网络(Temporal Convolutional Network,TCN)的时序处理能力和长短期记忆(Long Short Term Memory,LSTM)神经网络的门控记忆机制,构建了基于TCNLSTM网络的车辆换道意图识别模型。首先,将目标车辆的驾驶意图分为直行、向左换道和向右换道3种类型,从CitySim车辆轨迹数据集中提取出目标车辆及对应同车道、左侧车道、右侧车道的相邻前车和相邻后车的轨迹数据,并利用中值滤波算法获得车辆运行状态指标。其次,针对统计学理论和机器学习方法面临的识别精度不高、训练时间长、参数更新慢等问题,提出利用膨胀卷积技术提取时间序列的时序特征,采用门控记忆单元捕捉时序特征的长期依赖关系,并以目标车辆及周边相邻车辆的速度、加速度、航向角、航向角变化率和相对位置信息等54个车辆状态指标为输入变量,以车辆的换道意图为输出变量,构建了一个基于TCN-LSTM网络的车辆换道意图识别模型。最后,对比分析了不同输入时间步长下TCN、支持向量机(Support Vector Machines,SVM)、LSTM和TCN-LSTM模型的识别精度。结果表明:输入时间序列长度为150帧时,TCN-LSTM模型的识别精度达到最高值96.67%;从整体分类精度来看,相比LSTM、TCN和SVM模型,TCN-LSTM模型的换道意图分类准确率分别提升了1.34、0.84和2.46个百分点,展现出了更高的分类性能。展开更多
Heavily congested intersections in metropolitan areas in China are facing unique problems due to high travel demand and a high degree of traffic law violations. Based on a study conducted by the authors of this paper,...Heavily congested intersections in metropolitan areas in China are facing unique problems due to high travel demand and a high degree of traffic law violations. Based on a study conducted by the authors of this paper, 93% of left-turn vehicles turning left in these areas were slowed in order to avoid conflict with pedestrians. Intertwined pedestrian and vehicular flows can significantly reduce the capacity of exclusive left-turn lane group through reducing saturation flow rate, which increases the congestion at intersections. This paper investigates how the saturation flow rate of exclusive left-turn lane group is affected by the characteristics of pedestrian flow. By analyzing the imagery data collected by video cameras installed at intersections, the research team is able to obtain the characteristics of both vehicular and pedestrian flows, such as speed and spatial locations. The average operating speed at the saturation flow rate with and without pedestrian traffic is used as a direct measurement to evaluate the effect of pedestrians. Based on the statistical analysis, the paper concludes that saturation flow rate is mainly affected by the position of pedestrian in the crosswalk (inside or outside of left-turn vehicle’s trajectory), and the distance between the vehicle and pedestrians. In general, when the distance is less than four meters, the smaller the distance between vehicle and pedestrians, the larger the impact. However, there is no significant impact when the distance is larger than four meters. To accurately quantify the effect, the degree of pedestrian-vehicle impact is defined in four levels. The results show that the difference in the saturation flow rate between the best and the worst level could be 15.7%, which clearly indicates how important it is to enforce pedestrian crossing behavior.展开更多
文摘为及时识别、预测车辆的换道行为,综合考虑目标车辆及周边车辆的时空交互关系,结合时间卷积网络(Temporal Convolutional Network,TCN)的时序处理能力和长短期记忆(Long Short Term Memory,LSTM)神经网络的门控记忆机制,构建了基于TCNLSTM网络的车辆换道意图识别模型。首先,将目标车辆的驾驶意图分为直行、向左换道和向右换道3种类型,从CitySim车辆轨迹数据集中提取出目标车辆及对应同车道、左侧车道、右侧车道的相邻前车和相邻后车的轨迹数据,并利用中值滤波算法获得车辆运行状态指标。其次,针对统计学理论和机器学习方法面临的识别精度不高、训练时间长、参数更新慢等问题,提出利用膨胀卷积技术提取时间序列的时序特征,采用门控记忆单元捕捉时序特征的长期依赖关系,并以目标车辆及周边相邻车辆的速度、加速度、航向角、航向角变化率和相对位置信息等54个车辆状态指标为输入变量,以车辆的换道意图为输出变量,构建了一个基于TCN-LSTM网络的车辆换道意图识别模型。最后,对比分析了不同输入时间步长下TCN、支持向量机(Support Vector Machines,SVM)、LSTM和TCN-LSTM模型的识别精度。结果表明:输入时间序列长度为150帧时,TCN-LSTM模型的识别精度达到最高值96.67%;从整体分类精度来看,相比LSTM、TCN和SVM模型,TCN-LSTM模型的换道意图分类准确率分别提升了1.34、0.84和2.46个百分点,展现出了更高的分类性能。
文摘Heavily congested intersections in metropolitan areas in China are facing unique problems due to high travel demand and a high degree of traffic law violations. Based on a study conducted by the authors of this paper, 93% of left-turn vehicles turning left in these areas were slowed in order to avoid conflict with pedestrians. Intertwined pedestrian and vehicular flows can significantly reduce the capacity of exclusive left-turn lane group through reducing saturation flow rate, which increases the congestion at intersections. This paper investigates how the saturation flow rate of exclusive left-turn lane group is affected by the characteristics of pedestrian flow. By analyzing the imagery data collected by video cameras installed at intersections, the research team is able to obtain the characteristics of both vehicular and pedestrian flows, such as speed and spatial locations. The average operating speed at the saturation flow rate with and without pedestrian traffic is used as a direct measurement to evaluate the effect of pedestrians. Based on the statistical analysis, the paper concludes that saturation flow rate is mainly affected by the position of pedestrian in the crosswalk (inside or outside of left-turn vehicle’s trajectory), and the distance between the vehicle and pedestrians. In general, when the distance is less than four meters, the smaller the distance between vehicle and pedestrians, the larger the impact. However, there is no significant impact when the distance is larger than four meters. To accurately quantify the effect, the degree of pedestrian-vehicle impact is defined in four levels. The results show that the difference in the saturation flow rate between the best and the worst level could be 15.7%, which clearly indicates how important it is to enforce pedestrian crossing behavior.
