AI's Limited Role in Finding Missing Flood Victims: Collaboration Over Accuracy

Strategic Insights -

The aftermath of a flood presents significant challenges to search and rescue teams, including the difficulty of spotting victims in the vast and complex terrain. While drones have become a standard tool for first responders, their aerial imagery requires significant scanning time, making it essential to identify areas of interest quickly. Artificial intelligence (AI) has been proposed as a potential solution, but current implementations fall short due to high error rates and challenges in classification. However, AI can still play a valuable role when combined with human collaboration, helping to narrow down and prioritize imagery for further inspection.

Key Takeaways:

  • Current AI systems have high error rates in identifying potential victims in flood imagery, often due to visual challenges and lack of training data.
  • Machine learning systems require significant computational power and training data to identify signs of victims in aerial imagery.
  • Existing classifiers are prone to overestimating the number of candidate locations, leading to overload on search-and-rescue teams.
  • Oblique views and lack of precise GPS location in drone images increase the time required for teams to search for victims.
  • AI can still contribute to search efforts by identifying clumps of debris large enough to contain remains or finding signs of artificial colors and construction debris associated with remains.

Statistics:

  • A single 20-minute drone flight can produce over 800 high-resolution images.
  • A group of "squinters" would require over 22 hours of effort to manually inspect 10 flights' worth of images.
  • Current AI systems have a high error rate in identifying potential victims in flood imagery.
  • Over 8,000 images can be produced from 10 flights, making manual inspection extremely time-consuming.
  • The ideal solution is an AI system that scans the entire image, prioritizes images with the strongest signs of victims, and highlights the area of the image for a responder to inspect.

Sources:

  • [1The Conversation -- USA -- By Robin R. Murphy, Professor of Computer Science and Engineering, Texas A&M University]
  • [The Conversation -- USA -- By Robin R. Murphy, Professor of Computer Science and Engineering, Texas A&M University]
  • [The Conversation -- USA -- By Robin R. Murphy, Professor of Computer Science and Engineering, Texas A&M University]