C2A Dataset: Human Detection in Disaster Scenarios

In the aftermath of natural disasters, accurate and timely human detection is critical for effective search and rescue missions. UAVs have become essential tools in disaster response, but their effectiveness has been limited by the availability of specialized datasets. The C2A dataset was developed to address this challenge, providing high-resolution images with real disaster backgrounds and a wide range of human poses, making it ideal for training AI systems focused on human detection in emergency environments.

Key Features:

• Total Images: 10,215 high-resolution images
• Human Instances: Over 360,000 annotated human instances
• Human Poses: Includes 5 pose categories:
  • Bent
  • Kneeling
  • Lying
  • Sitting
  • Upright
• Disaster Scenario Types:
  • Fire/Smoke
  • Flood
  • Collapsed Buildings/Rubble
  • Traffic Accidents
• Image Resolutions: Ranges from 123×152 to 5184×3456 pixels
• Annotations: Bounding box annotations for each human instance
• Sources:
  • Disaster Backgrounds: AIDER (Aerial Image Dataset for Emergency Response Applications)
  • Human Poses: LSP/MPII-MPHB (Multiple Poses Human Body)

Dataset Inspiration: The C2A dataset was created to advance AI-assisted search and rescue operations by providing a diverse set of disaster scenarios and human poses. It aims to:

• Improve human detection algorithms in complex and varied environments.
• Enhance model generalization across different disaster types and human poses.
• Support the development of AI technologies to assist first responders in saving lives.