MAY 31, 2024 by Bob Yirka , Tech Xplore
Collected at : https://techxplore.com/news/2024-05-ai-drones-lost-hikers.html
A trio of engineers at the University of Glasgow has designed and built an AI-based drone system that can assist in search efforts for hikers lost in the wilderness. Jan-Hendrik Ewers, David Anderson, and Douglas Thomson have published a paper describing their efforts on the arXiv preprint server.
Hiking in the Scottish Highlands has become a popular activity over the past several decades—the rugged landscape offers a wide variety of vistas and the remote locations allow hikers to get back to nature. But such hiking can be hazardous—hikers can get lost, or injured. Many find themselves in need of assistance every year. Because of that, emergency teams use both traditional and modern techniques to find those who are lost or have become disabled for some reason.
In recent years, searchers have begun to use drones—their higher vantage point allows for covering more land more quickly than searching on foot. It also helps spot things that searchers may not be able to see from the ground, such as trodden brush. In this new effort, the team at the University of Glasgow wondered if adding AI to drone technology could improve the search for lost hikers.
To find out, they created an AI model using data sets showing the paths taken by people who were lost and then found by search parties around the world. They also added pertinent data, such as age, reason for hiking, whether they were alone, walking, or on horseback or some other type of transport.
They noted geographical information regarding both the path taken by those who were lost and where they were found, such as rivers, streams, roads or open ground. They then added data specific to Scotland’s geography. They ran the model millions of times, each representing a simulated search, until it narrowed down the most probable paths a lost hiker would take. The drone would then be instructed to search those paths first.
In testing their system against traditional approaches—such as using “the lawn mower” sweeping technique—used to find actual hikers, the new approach found lost hikers more often—19% of the time as compared to 8–12% of the time. The researchers suggest more data should make their system more accurate and eventually result in a tool that could be used to save lives.
More information: Jan-Hendrik Ewers et al, Deep Reinforcement Learning for Time-Critical Wilderness Search And Rescue Using Drones, arXiv (2024). DOI: 10.48550/arxiv.2405.12800
Journal information: arXiv
Leave a Reply