高速公路匝道汇入场景识别与分析

Recognition and Analysis of Highway Entrance Ramp Merging Scenarios

摘要:

目前关于高速公路入口匝道汇入行为的划分方法仅仅局限于单次单一交互的驾驶场景，而未考虑换道车辆与目标车道周围车辆时空位置的动态变化，即忽视了对双向多交互汇入场景的研究，这对于研究者把握真实的交通流特性，开发自动驾驶系统辅助换道算法带来了巨大挑战。为了揭示实际汇入过程中自然驾驶车辆间的动态交互行为，通过ExiD数据集与高精地图数据的结合进行自然驾驶汇入轨迹数据的提取，建立能充分考虑时空动态变化的汇入场景划分方法，并利用詹森?香农散度构建场景相似性量化模型与指标，以表征不同汇入场景的差异性与相似性。结果表明，在考虑目标车道周围车辆的时空动态分布后，将匝道汇入行为细分为8种具有显著差异性的场景，甚至包括具有较高风险的存在着并行车辆的场景（平均时距小于2 s），这对于解析高风险汇入场景以及设计出符合人类驾驶人交互特征的自动驾驶汇入策略具有重要的参考价值。

Abstract:

The current classification methods for highway entrance ramp merging behavior are limited to single and isolated driving scenarios， without considering the dynamic changes in the spatial and temporal positions of merging vehicles and surrounding vehicles in the target lane， neglecting the study of bidirectional and multi-interaction merging scenarios， which poses significant challenges for researchers in understanding real traffic flow characteristics and developing automated driving systems with lane-changing algorithms. To reveal the dynamic interactive behavior among vehicles during actual merging processes， this paper， combining the ExiD dataset with high-definition map data to extract natural driving merging trajectory data， proposes a merging scenario classification method that fully considers the spatiotemporal dynamic changes. Additionally， it utilizes the Jensen-Shannon divergence to construct a quantitative model and index for measuring scenario similarity， representing the differences and similarities among different merging scenarios. The results show that， after considering the spatiotemporal distribution of surrounding vehicles in the target lane， the merging behavior on the ramp can be divided into eight distinct scenarios with significant differences， including scenarios with higher risks， such as parallel vehicles （average time-to-collision 2 s）. These findings are of great importance in analyzing high-risk merging scenarios and designing automated merging strategies that align with the interaction characteristics of human drivers.