Why drones and AI can’t quickly find missing flood victims, yet

Source Domain: www.newstimes.com

Summary of the Article on AI in Search and Rescue:

AI Speed: AI systems, specifically through computer vision and machine learning, significantly outpace human responders in scanning high-resolution drone imagery, often completing this task in under a second compared to one to three minutes for humans.
AI Limitations: Despite their speed, current AI systems are not accurate enough to replace human search-and-rescue teams in finding flood victims. This includes issues like high error rates in detection due to the unique and complex imagery created by flood conditions.
Specialized Challenges: Flood victims often present distinct visual challenges to AI, including being obscured by debris, water, and mud, which differ substantially from the indicators used for more straightforward search-and-rescue situations.
Lack of Training Data: There is a shortage of training datasets containing aerial imagery of individuals tangled in flood-related debris, complicating the development of precise machine learning classifiers.
Geographical Inaccuracies: Drones often capture imagery at oblique angles, leading to imprecise GPS coordinates for targeted search areas, which requires additional manual verification.
AI-Human Collaboration: While AI systems alone are currently insufficient, combining them with human search-and-rescue teams offers better search efficiency by prioritizing potentially critical areas and aiding faster response.
Future Potential: With improvements in training data and algorithm accuracy, AI could eventually play a more significant role in helping to identify debris and human signs, especially in early search stages and later verifying no areas were overlooked.