DACT (Dataset of Annotated Car Trajectories)

DACT包含注释的汽车轨迹数据的两个子集。该数据集包含50条轨迹，涵盖了大约13个小时的驾驶数据。在DACT中，我们使用交互式框架手动指定了重要的驾驶模式。重要的驾驶模式可以是转弯，加速，减速等。注释过程由众包任务组成，然后是全面的聚合阶段。聚合是通过两种不同的策略完成的: 严格和容易。对于第一个，我们使用了一些严格的约束来汇总众包结果，而我们使用了灵活的约束来生成DACT的第二个子集。

DACT contains two subsets of annotated vehicle trajectory data. The dataset comprises 50 trajectories spanning approximately 13 hours of driving recordings. In DACT, we manually designated key driving maneuvers via an interactive framework; these maneuvers may include turns, acceleration, deceleration, and other similar driving actions. The annotation workflow consists of crowdsourcing tasks followed by a comprehensive aggregation phase. Aggregation is implemented using two distinct strategies: strict and easy. For the strict strategy, we applied rigorous constraints to aggregate the crowdsourced annotations, whereas flexible constraints were employed to generate the second subset of DACT.

搜集汇总

背景与挑战

背景概述

DACT数据集包含两个子集，涵盖50条轨迹和约13小时的驾驶数据，通过手动标注了重要驾驶模式如转弯和加速。标注过程涉及众包任务和聚合阶段，聚合策略分为严格和容易两种。