基于智慧路口对抗交通流的合成数据集

基于智慧路口对抗交通流的合成数据集，依托路侧数据提取真实的交通流信息，对局部场景进行重建和全局场景进行融合，实现高精度真实动态场景的重建，随后基于TTC和PET等指标利用场景挖掘算法对变道、制动、典型交通流功能场景进行提取，形成真实场景库；基于真实场景，使用基于AI的自然及对抗交通流模型进行数据合成，通过城区交叉路口交通流场景合成技术，对原始单一场景进行泛化，增强自动驾驶训练数据集的多样性、解决仿真测试极端场景稀疏性，生成高质量真实路侧场景、自然合成场景、对抗合成场景的多样极端驾驶场景，总共60个场景数据集，遵循OpenX标准，以xosc格式文件描述动态交通数据，包含基本道路信息以及不同类别交通参与者的运动轨迹。

A synthetic dataset for adversarial traffic flows at intelligent intersections. It extracts real traffic flow information from roadside data, reconstructs local scenarios and fuses global scenarios to achieve high-fidelity reconstruction of real dynamic traffic scenarios. Subsequently, scenario mining algorithms are used to extract lane-changing, braking, and typical traffic flow functional scenarios based on metrics such as TTC and PET, forming a real scenario library. Based on these real scenarios, AI-powered natural and adversarial traffic flow models are applied for data synthesis. Through urban intersection traffic flow scenario synthesis technology, the original single scenarios are generalized, enhancing the diversity of autonomous driving training datasets and addressing the sparsity of extreme scenarios in simulation testing. This generates diverse extreme driving scenarios including high-quality real roadside scenarios, naturally synthesized scenarios, and adversarial synthesized scenarios. There are 60 scenario datasets in total, which follow the OpenX standard, use XOSC format files to describe dynamic traffic data, and contain basic road information as well as motion trajectories of various types of traffic participants.

搜集汇总

背景与挑战

背景概述

该数据集基于智慧路口的车路云一体化数据，通过AI交通流合成技术生成了60个遵循OpenX标准的对抗交通流场景，旨在增强自动驾驶训练数据的多样性和解决仿真测试中极端场景稀疏性问题。它覆盖了多种高交互场景和超过30%的Corner Case，支持主流仿真器导入以实现高保真还原。