Scenario data play a vital role in autonomous driving related researches, and it is essential to obtain refined descriptions and labels to extract and index scenarios with different types of interactions. However, existing methods cannot cope well with the problem of scenario classification and comparison with vehicle interactions as the core. In this paper, we propose a framework for interaction-based refined scenario classification and labeling. Based on the summarized basic types of vehicle interactions, we slice scenario data stream into a series of scenario segments via spatiotemporal scenario evolution tree. The scenario segment statistics of many published scenario datasets are further analyzed. We also propose the scenario metric Graph-DTW based on Graph Computation Tree and Dynamic Time Warping to conduct refined scenario comparison and labeling. The extreme interactive scenarios and corner cases can be efficiently filtered and extracted. Moreover, testing examples on trajectory prediction model demonstrate the effectiveness and advantages of scenario labeling and the proposed metric. The overall framework can provide solid support for the usage and indexing of scenario data.

翻译：场景数据在自动驾驶相关研究中至关重要，获取精细化的描述和标注以提取和索引具有不同交互类型的场景十分必要。然而，现有方法难以有效解决以车辆交互为核心的场景分类与比较问题。本文提出了一种基于交互的精细化场景分类与标注框架。基于归纳的车辆交互基本类型，我们通过时空场景演化树将场景数据流切分为一系列场景片段。进一步分析了多个公开场景数据集的场景片段统计特征。同时，我们提出基于图计算树与动态时间规整的场景度量指标Graph-DTW，用于实现精细化场景比较与标注。该方法可高效筛选和提取极端交互场景与边界案例。最后，在轨迹预测模型上的测试实例验证了场景标注及所提度量的有效性与优势。该整体框架可为场景数据的使用与索引提供坚实支撑。