Scholarlink

The development of autonomous driving systems increasingly relies on rich, diverse, and well-annotated datasets, particularly for training deep learning models. However, collecting real-world data, especially from event cameras, remains a costly and labor-intensive task. In this work, we introduce a synthetic data generation tool built upon the CARLA simulator and extended with advanced event simulation capabilities via the V2E (Video to Events) framework. Our system enables the simulation of driving scenarios with fine-grained control over weather, lighting, traffic, and sensor configurations. It supports multi-modal sensor data capture, including RGB, depth, semantic segmentation, and instance segmentation, as well as realistic DVS event streams. By combining CARLA’s diverse environments and rich sensor suite with high-fidelity event simulation, the tool facilitates the creation of scalable, customizable datasets that are well-suited for training and evaluating both frame-based and event-based perception algorithms. Code:

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Event-Based Data Generation Framework for Autonomous Driving Scenarios

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Abdessamad El Kaouri,
  • Mohamed Kas,
  • Yassine Ruichek,
  • Youssef El Merabet

摘要

The development of autonomous driving systems increasingly relies on rich, diverse, and well-annotated datasets, particularly for training deep learning models. However, collecting real-world data, especially from event cameras, remains a costly and labor-intensive task. In this work, we introduce a synthetic data generation tool built upon the CARLA simulator and extended with advanced event simulation capabilities via the V2E (Video to Events) framework. Our system enables the simulation of driving scenarios with fine-grained control over weather, lighting, traffic, and sensor configurations. It supports multi-modal sensor data capture, including RGB, depth, semantic segmentation, and instance segmentation, as well as realistic DVS event streams. By combining CARLA’s diverse environments and rich sensor suite with high-fidelity event simulation, the tool facilitates the creation of scalable, customizable datasets that are well-suited for training and evaluating both frame-based and event-based perception algorithms. Code: