Autonomous Driving Test Scenario Generation and Integration Using UAV-Based Visual Data

Engineering notes

Key topics: autonomous driving, semantic segmentation, multi-object tracking, object tracking, object detection. See the paper for implementation details and experimental results.

Chinese explanation / 中文解读

中文解读待补充：本站会优先为端到端自动驾驶、BEV感知、3D目标检测、轨迹预测、路径规划、LiDAR感知等高价值论文补充中文说明。

Original abstract

To efficiently and realistically validate autonomous driving functions, this paper proposes a framework that automatically generates and integrates test scenarios using visual data captured by unmanned aerial vehicles (UAVs). This approach overcomes the limitations of manual scene construction by extracting static and dynamic elements from real-world traffic videos. Static features, such as road geometry and lane topology, are obtained through semantic segmentation, while dynamic elements, including vehicle trajectories, are obtained through object detection and multi-object tracking. These elements are formalized as ASAM OpenDRIVE, and the structured outputs are imported into a simulation environment for validation. Experimental results demonstrate that the framework can reproduce real-world scenarios and behaviours with high fidelity, providing a scalable solution for scenario-based testing in the development of autonomous driving.

Links and sources

• Official / arXiv page

Comments