Core Concepts
Recent advancements in V2X technologies have led to the creation of the V2X-Real dataset, facilitating cooperative perception research.
Abstract
Introduction:
Recent advancements in V2X tech enable autonomous vehicles to share sensing info.
Lack of real-world datasets for V2X cooperative perception research.
Data Acquisition:
Dataset collected using two connected automated vehicles and smart infrastructures.
Contains LiDAR frames, camera data, and annotated bounding boxes.
Data Annotation and Processing:
Expert annotators used SUSTechPOINTS tool for 3D bounding box annotations.
Strategies implemented to enhance annotation efficiency and accuracy.
Dataset Analysis:
Average of 36 objects per scene with high traffic density in urban scenarios.
Distribution of annotations across categories like pedestrian, car, truck shown.
Tasks:
Four sub-datasets derived for different collaboration modes: VC, IC, V2V, I2I.
Metrics:
Evaluation conducted based on AP under IoU thresholds for multi-class detection tasks.
Stats
この論文では、33KのLiDARフレーム、171Kのカメラ画像、および1.2M以上のアノテーション付き3Dバウンディングボックスを含むV2X協力知覚用の大規模な実世界データセットが紹介されています。