Could artificial intelligence play a role in rescuing people from drowning? SFI is funding a project at NUI Galway that’s using game-based technology to train AI to spot humans in water, and the researchers hope it could help to spot humans who are at risk of drowning or aid retrieval missions when someone has been swept away.

The ALIVE (Autonomous LIfeguard and Vision Environment) project is developing a camera-based system to identify where and when people are at risk of drowning and alert emergency personnel. The research is inspired by a strong and often local need, according to Dr Enda Barrett, who works on the project with Professor Michael Madden in the School of Computer Science and Insight Data Analytics SFI Research Centre at NUI Galway, “The Corrib is a fast-flowing river here in Galway and sadly there are several drownings in the river and in nearby locations, such as Salthill. Volunteers patrol the more dangerous spots, in a bid to prevent people from entering the water.”

The ALIVE system could offer a more permanent monitoring approach, using computer vision and deep learning techniques applied cameras stationed around the water to ‘see’ if a human is in trouble and send out the alert. Dr Barrett explains, “We are developing it using simple cameras, nothing particularly expensive or special, and we want to train the software to detect a body going into the water, or track it in the water.”

The key though, is to ensure that the AI system knows a human body from all sorts of other objects, “In this environment you have waves, seagulls, reflections, all sorts of things that make it noisy from the computer’s point of view, and what we don’t want is the system alerting emergency services when they are not needed.”

In order to learn what a human body looks like when falling into or moving through water, the AI needs to be shown many thousands of such images, but these are difficult to source. Enter gaming technology: the researchers have built 3D virtual environment simulations to mimic what happens in such events. “The machine needs to make sense of pixels and classify if there is a person in the frame, and the simulation allows us to teach the machine what that looks like under many different conditions,” says Dr Barrett.

Ultimately, he would like to see such a system set up along rivers and shorelines to assist humans in distress by sending for help - possibly even deploying drones with buoyancy aids to the person. Or, if the person has already drowned, the system could provide information about where they entered the water and this would help with retrieval missions. “There is a lot for the system to learn, but we hope that one day it could save lives and reduce the risks and costs of searches."